Professional Documents
Culture Documents
Guide Tuning Pro
Guide Tuning Pro
E85762-04
Copyright © 2013, 2019, Oracle and/or its affiliates. All rights reserved.
Contributors: Hermann Baer, Ali Cakmak, Sunil Chakkappen, Immanuel Chan, Deba Chatterjee, Chris
Chiappa, Dinesh Das, Leonidas Galanis, William Endress, Marcus Fallen, Bruce Golbus, Katsumi Inoue,
Shantanu Joshi, Adam Kociubes, Keith Laker, Allison Lee, Sue Lee, David McDermid, Colin McGregor, Ajit
Mylavarapu, Ted Persky, Lei Sheng, Ekrem Soylemez, Hong Su, Murali Thiyagarajah, Randy Urbano, Sahil
Vazirani, Bharath Venkatakrishnan, Hailing Yu, John Zimmerman
This software and related documentation are provided under a license agreement containing restrictions on
use and disclosure and are protected by intellectual property laws. Except as expressly permitted in your
license agreement or allowed by law, you may not use, copy, reproduce, translate, broadcast, modify,
license, transmit, distribute, exhibit, perform, publish, or display any part, in any form, or by any means.
Reverse engineering, disassembly, or decompilation of this software, unless required by law for
interoperability, is prohibited.
The information contained herein is subject to change without notice and is not warranted to be error-free. If
you find any errors, please report them to us in writing.
If this is software or related documentation that is delivered to the U.S. Government or anyone licensing it on
behalf of the U.S. Government, then the following notice is applicable:
U.S. GOVERNMENT END USERS: Oracle programs, including any operating system, integrated software,
any programs installed on the hardware, and/or documentation, delivered to U.S. Government end users are
"commercial computer software" pursuant to the applicable Federal Acquisition Regulation and agency-
specific supplemental regulations. As such, use, duplication, disclosure, modification, and adaptation of the
programs, including any operating system, integrated software, any programs installed on the hardware,
and/or documentation, shall be subject to license terms and license restrictions applicable to the programs.
No other rights are granted to the U.S. Government.
This software or hardware is developed for general use in a variety of information management applications.
It is not developed or intended for use in any inherently dangerous applications, including applications that
may create a risk of personal injury. If you use this software or hardware in dangerous applications, then you
shall be responsible to take all appropriate fail-safe, backup, redundancy, and other measures to ensure its
safe use. Oracle Corporation and its affiliates disclaim any liability for any damages caused by use of this
software or hardware in dangerous applications.
Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of
their respective owners.
Intel and Intel Xeon are trademarks or registered trademarks of Intel Corporation. All SPARC trademarks are
used under license and are trademarks or registered trademarks of SPARC International, Inc. AMD, Opteron,
the AMD logo, and the AMD Opteron logo are trademarks or registered trademarks of Advanced Micro
Devices. UNIX is a registered trademark of The Open Group.
This software or hardware and documentation may provide access to or information about content, products,
and services from third parties. Oracle Corporation and its affiliates are not responsible for and expressly
disclaim all warranties of any kind with respect to third-party content, products, and services unless otherwise
set forth in an applicable agreement between you and Oracle. Oracle Corporation and its affiliates will not be
responsible for any loss, costs, or damages incurred due to your access to or use of third-party content,
products, or services, except as set forth in an applicable agreement between you and Oracle.
Contents
Preface
Audience xxv
Documentation Accessibility xxv
Related Documents xxvi
Conventions xxvi
iii
2.2.1 Guideline for Deploying in a Test Environment 2-3
2.2.2 Guidelines for Application Rollout 2-4
3 SQL Processing
3.1 About SQL Processing 3-1
3.1.1 SQL Parsing 3-3
3.1.1.1 Syntax Check 3-3
3.1.1.2 Semantic Check 3-4
3.1.1.3 Shared Pool Check 3-4
3.1.2 SQL Optimization 3-6
3.1.3 SQL Row Source Generation 3-7
3.1.4 SQL Execution 3-8
3.2 How Oracle Database Processes DML 3-10
3.2.1 How Row Sets Are Fetched 3-10
3.2.2 Read Consistency 3-11
3.2.3 Data Changes 3-11
3.3 How Oracle Database Processes DDL 3-11
iv
4.4.1.2 Purpose of Adaptive Query Plans 4-16
4.4.1.3 How Adaptive Query Plans Work 4-16
4.4.1.4 When Adaptive Query Plans Are Enabled 4-23
4.4.2 Adaptive Statistics 4-24
4.4.2.1 Dynamic Statistics 4-24
4.4.2.2 Automatic Reoptimization 4-24
4.4.2.3 SQL Plan Directives 4-28
4.4.2.4 When Adaptive Statistics Are Enabled 4-29
4.5 About Approximate Query Processing 4-29
4.6 About SQL Plan Management 4-31
4.7 About the Expression Statistics Store (ESS) 4-32
5 Query Transformations
5.1 OR Expansion 5-2
5.2 View Merging 5-4
5.2.1 Query Blocks in View Merging 5-5
5.2.2 Simple View Merging 5-5
5.2.3 Complex View Merging 5-8
5.3 Predicate Pushing 5-11
5.4 Subquery Unnesting 5-12
5.5 Query Rewrite with Materialized Views 5-12
5.6 Star Transformation 5-13
5.6.1 About Star Schemas 5-13
5.6.2 Purpose of Star Transformations 5-14
5.6.3 How Star Transformation Works 5-14
5.6.4 Controls for Star Transformation 5-15
5.6.5 Star Transformation: Scenario 5-15
5.6.6 Temporary Table Transformation: Scenario 5-18
5.7 In-Memory Aggregation (VECTOR GROUP BY) 5-20
5.8 Cursor-Duration Temporary Tables 5-21
5.8.1 Purpose of Cursor-Duration Temporary Tables 5-21
5.8.2 How Cursor-Duration Temporary Tables Work 5-21
5.8.3 Cursor-Duration Temporary Tables: Example 5-22
5.9 Table Expansion 5-23
5.9.1 Purpose of Table Expansion 5-24
5.9.2 How Table Expansion Works 5-24
5.9.3 Table Expansion: Scenario 5-25
5.9.4 Table Expansion and Star Transformation: Scenario 5-29
5.10 Join Factorization 5-31
5.10.1 Purpose of Join Factorization 5-32
v
5.10.2 How Join Factorization Works 5-32
5.10.3 Factorization and Join Orders: Scenario 5-33
5.10.4 Factorization of Outer Joins: Scenario 5-34
vi
7.2.5.2 Pruning Information with Composite Partitioned Objects:
Examples 7-14
7.2.5.3 Examples of Partial Partition-Wise Joins 7-16
7.2.5.4 Example of Full Partition-Wise Join 7-18
7.2.5.5 Examples of INLIST ITERATOR and EXPLAIN PLAN 7-19
7.2.5.6 Example of Domain Indexes and EXPLAIN PLAN 7-21
7.2.6 PLAN_TABLE Columns 7-21
7.3 Execution Plan Reference 7-32
7.3.1 Execution Plan Views 7-32
7.3.2 PLAN_TABLE Columns 7-33
7.3.3 DBMS_XPLAN Display Functions 7-43
vii
8.3.1.3 Unique and Nonunique Indexes 8-18
8.3.1.4 B-Tree Indexes and Nulls 8-18
8.3.2 Index Unique Scans 8-21
8.3.2.1 When the Optimizer Considers Index Unique Scans 8-21
8.3.2.2 How Index Unique Scans Work 8-22
8.3.2.3 Index Unique Scans: Example 8-23
8.3.3 Index Range Scans 8-24
8.3.3.1 When the Optimizer Considers Index Range Scans 8-24
8.3.3.2 How Index Range Scans Work 8-25
8.3.3.3 Index Range Scan: Example 8-27
8.3.3.4 Index Range Scan Descending: Example 8-27
8.3.4 Index Full Scans 8-28
8.3.4.1 When the Optimizer Considers Index Full Scans 8-28
8.3.4.2 How Index Full Scans Work 8-29
8.3.4.3 Index Full Scans: Example 8-30
8.3.5 Index Fast Full Scans 8-30
8.3.5.1 When the Optimizer Considers Index Fast Full Scans 8-31
8.3.5.2 How Index Fast Full Scans Work 8-31
8.3.5.3 Index Fast Full Scans: Example 8-31
8.3.6 Index Skip Scans 8-32
8.3.6.1 When the Optimizer Considers Index Skips Scans 8-32
8.3.6.2 How Index Skip Scans Work 8-32
8.3.6.3 Index Skip Scans: Example 8-33
8.3.7 Index Join Scans 8-34
8.3.7.1 When the Optimizer Considers Index Join Scans 8-35
8.3.7.2 How Index Join Scans Work 8-35
8.3.7.3 Index Join Scans: Example 8-35
8.4 Bitmap Index Access Paths 8-37
8.4.1 About Bitmap Index Access 8-37
8.4.1.1 Differences Between Bitmap and B-Tree Indexes 8-38
8.4.1.2 Purpose of Bitmap Indexes 8-38
8.4.1.3 Bitmaps and Rowids 8-40
8.4.1.4 Bitmap Join Indexes 8-41
8.4.1.5 Bitmap Storage 8-42
8.4.2 Bitmap Conversion to Rowid 8-42
8.4.2.1 When the Optimizer Chooses Bitmap Conversion to Rowid 8-43
8.4.2.2 How Bitmap Conversion to Rowid Works 8-43
8.4.2.3 Bitmap Conversion to Rowid: Example 8-43
8.4.3 Bitmap Index Single Value 8-44
8.4.3.1 When the Optimizer Considers Bitmap Index Single Value 8-44
8.4.3.2 How Bitmap Index Single Value Works 8-44
viii
8.4.3.3 Bitmap Index Single Value: Example 8-44
8.4.4 Bitmap Index Range Scans 8-45
8.4.4.1 When the Optimizer Considers Bitmap Index Range Scans 8-46
8.4.4.2 How Bitmap Index Range Scans Work 8-46
8.4.4.3 Bitmap Index Range Scans: Example 8-46
8.4.5 Bitmap Merge 8-47
8.4.5.1 When the Optimizer Considers Bitmap Merge 8-47
8.4.5.2 How Bitmap Merge Works 8-48
8.4.5.3 Bitmap Merge: Example 8-48
8.5 Table Cluster Access Paths 8-49
8.5.1 Cluster Scans 8-49
8.5.1.1 When the Optimizer Considers Cluster Scans 8-50
8.5.1.2 How a Cluster Scan Works 8-50
8.5.1.3 Cluster Scans: Example 8-50
8.5.2 Hash Scans 8-51
8.5.2.1 When the Optimizer Considers a Hash Scan 8-52
8.5.2.2 How a Hash Scan Works 8-52
8.5.2.3 Hash Scans: Example 8-52
9 Joins
9.1 About Joins 9-1
9.1.1 Join Trees 9-2
9.1.2 How the Optimizer Executes Join Statements 9-3
9.1.3 How the Optimizer Chooses Execution Plans for Joins 9-4
9.2 Join Methods 9-5
9.2.1 Nested Loops Joins 9-6
9.2.1.1 When the Optimizer Considers Nested Loops Joins 9-7
9.2.1.2 How Nested Loops Joins Work 9-7
9.2.1.3 Nested Nested Loops 9-8
9.2.1.4 Current Implementation for Nested Loops Joins 9-11
9.2.1.5 Original Implementation for Nested Loops Joins 9-14
9.2.1.6 Nested Loops Controls 9-15
9.2.2 Hash Joins 9-17
9.2.2.1 When the Optimizer Considers Hash Joins 9-18
9.2.2.2 How Hash Joins Work 9-18
9.2.2.3 How Hash Joins Work When the Hash Table Does Not Fit in the
PGA 9-21
9.2.2.4 Hash Join Controls 9-22
9.2.3 Sort Merge Joins 9-22
9.2.3.1 When the Optimizer Considers Sort Merge Joins 9-23
9.2.3.2 How Sort Merge Joins Work 9-23
ix
9.2.3.3 Sort Merge Join Controls 9-27
9.3 Join Types 9-28
9.3.1 Inner Joins 9-28
9.3.1.1 Equijoins 9-29
9.3.1.2 Nonequijoins 9-29
9.3.1.3 Band Joins 9-30
9.3.2 Outer Joins 9-34
9.3.2.1 Nested Loops Outer Joins 9-34
9.3.2.2 Hash Join Outer Joins 9-35
9.3.2.3 Sort Merge Outer Joins 9-37
9.3.2.4 Full Outer Joins 9-37
9.3.2.5 Multiple Tables on the Left of an Outer Join 9-39
9.3.3 Semijoins 9-39
9.3.3.1 When the Optimizer Considers Semijoins 9-40
9.3.3.2 How Semijoins Work 9-40
9.3.4 Antijoins 9-42
9.3.4.1 When the Optimizer Considers Antijoins 9-43
9.3.4.2 How Antijoins Work 9-44
9.3.4.3 How Antijoins Handle Nulls 9-45
9.3.5 Cartesian Joins 9-48
9.3.5.1 When the Optimizer Considers Cartesian Joins 9-49
9.3.5.2 How Cartesian Joins Work 9-49
9.3.5.3 Cartesian Join Controls 9-50
9.4 Join Optimizations 9-51
9.4.1 Bloom Filters 9-52
9.4.1.1 Purpose of Bloom Filters 9-52
9.4.1.2 How Bloom Filters Work 9-52
9.4.1.3 Bloom Filter Controls 9-53
9.4.1.4 Bloom Filter Metadata 9-54
9.4.1.5 Bloom Filters: Scenario 9-54
9.4.2 Partition-Wise Joins 9-56
9.4.2.1 Purpose of Partition-Wise Joins 9-57
9.4.2.2 How Partition-Wise Joins Work 9-57
9.4.3 In-Memory Join Groups 9-60
x
10.2.1 Table Statistics 10-4
10.2.2 Column Statistics 10-5
10.2.3 Index Statistics 10-5
10.2.3.1 Types of Index Statistics 10-5
10.2.3.2 Index Clustering Factor 10-6
10.2.3.3 Effect of Index Clustering Factor on Cost: Example 10-10
10.2.4 Session-Specific Statistics for Global Temporary Tables 10-11
10.2.4.1 Shared and Session-Specific Statistics for Global Temporary
Tables 10-11
10.2.4.2 Effect of DBMS_STATS on Transaction-Specific Temporary
Tables 10-12
10.2.5 System Statistics 10-13
10.2.6 User-Defined Optimizer Statistics 10-13
10.3 How the Database Gathers Optimizer Statistics 10-14
10.3.1 DBMS_STATS Package 10-14
10.3.2 Supplemental Dynamic Statistics 10-15
10.3.3 Online Statistics Gathering for Bulk Loads 10-16
10.3.3.1 Purpose of Online Statistics Gathering for Bulk Loads 10-17
10.3.3.2 Global Statistics During Inserts into Empty Partitioned Tables 10-17
10.3.3.3 Index Statistics and Histograms During Bulk Loads 10-18
10.3.3.4 Restrictions for Online Statistics Gathering for Bulk Loads 10-18
10.3.3.5 Hints for Online Statistics Gathering for Bulk Loads 10-20
10.4 When the Database Gathers Optimizer Statistics 10-20
10.4.1 Sources for Optimizer Statistics 10-20
10.4.2 SQL Plan Directives 10-21
10.4.2.1 When the Database Creates SQL Plan Directives 10-22
10.4.2.2 How the Database Uses SQL Plan Directives 10-22
10.4.2.3 SQL Plan Directive Maintenance 10-23
10.4.2.4 How the Optimizer Uses SQL Plan Directives: Example 10-24
10.4.2.5 How the Optimizer Uses Extensions and SQL Plan Directives:
Example 10-29
10.4.3 When the Database Samples Data 10-33
10.4.4 How the Database Samples Data 10-35
11 Histograms
11.1 Purpose of Histograms 11-2
11.2 When Oracle Database Creates Histograms 11-2
11.3 How Oracle Database Chooses the Histogram Type 11-4
11.4 Cardinality Algorithms When Using Histograms 11-5
11.4.1 Endpoint Numbers and Values 11-5
11.4.2 Popular and Nonpopular Values 11-5
xi
11.4.3 Bucket Compression 11-6
11.5 Frequency Histograms 11-7
11.5.1 Criteria For Frequency Histograms 11-7
11.5.2 Generating a Frequency Histogram 11-8
11.6 Top Frequency Histograms 11-11
11.6.1 Criteria For Top Frequency Histograms 11-11
11.6.2 Generating a Top Frequency Histogram 11-12
11.7 Height-Balanced Histograms (Legacy) 11-15
11.7.1 Criteria for Height-Balanced Histograms 11-15
11.7.2 Generating a Height-Balanced Histogram 11-16
11.8 Hybrid Histograms 11-19
11.8.1 How Endpoint Repeat Counts Work 11-19
11.8.2 Criteria for Hybrid Histograms 11-21
11.8.3 Generating a Hybrid Histogram 11-22
xii
13.1.2 Configuring Automatic Optimizer Statistics Collection Using Cloud
Control 13-2
13.1.3 Configuring Automatic Optimizer Statistics Collection from the
Command Line 13-4
13.2 Gathering Optimizer Statistics Manually 13-6
13.2.1 About Manual Statistics Collection with DBMS_STATS 13-7
13.2.2 Guidelines for Gathering Optimizer Statistics Manually 13-8
13.2.2.1 Guideline for Setting the Sample Size 13-9
13.2.2.2 Guideline for Gathering Statistics in Parallel 13-10
13.2.2.3 Guideline for Partitioned Objects 13-10
13.2.2.4 Guideline for Frequently Changing Objects 13-11
13.2.2.5 Guideline for External Tables 13-11
13.2.3 Determining When Optimizer Statistics Are Stale 13-11
13.2.4 Gathering Schema and Table Statistics 13-13
13.2.5 Gathering Statistics for Fixed Objects 13-13
13.2.6 Gathering Statistics for Volatile Tables Using Dynamic Statistics 13-14
13.2.7 Gathering Optimizer Statistics Concurrently 13-16
13.2.7.1 About Concurrent Statistics Gathering 13-16
13.2.7.2 Enabling Concurrent Statistics Gathering 13-19
13.2.7.3 Monitoring Statistics Gathering Operations 13-22
13.2.8 Gathering Incremental Statistics on Partitioned Objects 13-24
13.2.8.1 Purpose of Incremental Statistics 13-24
13.2.8.2 How DBMS_STATS Derives Global Statistics for Partitioned
tables 13-25
13.2.8.3 Gathering Statistics for a Partitioned Table: Basic Steps 13-29
13.2.8.4 Maintaining Incremental Statistics for Partition Maintenance
Operations 13-32
13.2.8.5 Maintaining Incremental Statistics for Tables with Stale or
Locked Partition Statistics 13-34
13.3 Gathering System Statistics Manually 13-36
13.3.1 About Gathering System Statistics with DBMS_STATS 13-37
13.3.2 Guidelines for Gathering System Statistics 13-39
13.3.3 Gathering Workload Statistics 13-39
13.3.3.1 About Workload Statistics 13-39
13.3.3.2 Starting and Stopping System Statistics Gathering 13-41
13.3.3.3 Gathering System Statistics During a Specified Interval 13-42
13.3.4 Gathering Noworkload Statistics 13-43
13.3.5 Deleting System Statistics 13-45
13.4 Running Statistics Gathering Functions in Reporting Mode 13-45
xiii
14 Managing Extended Statistics
14.1 Managing Column Group Statistics 14-1
14.1.1 About Statistics on Column Groups 14-2
14.1.1.1 Why Column Group Statistics Are Needed: Example 14-3
14.1.1.2 Automatic and Manual Column Group Statistics 14-5
14.1.1.3 User Interface for Column Group Statistics 14-6
14.1.2 Detecting Useful Column Groups for a Specific Workload 14-7
14.1.3 Creating Column Groups Detected During Workload Monitoring 14-10
14.1.4 Creating and Gathering Statistics on Column Groups Manually 14-12
14.1.5 Displaying Column Group Information 14-13
14.1.6 Dropping a Column Group 14-14
14.2 Managing Expression Statistics 14-15
14.2.1 About Expression Statistics 14-15
14.2.1.1 When Expression Statistics Are Useful: Example 14-16
14.2.2 Creating Expression Statistics 14-17
14.2.3 Displaying Expression Statistics 14-18
14.2.4 Dropping Expression Statistics 14-19
xiv
16.2.2 Changing the Optimizer Statistics Retention Period 16-6
16.2.3 Purging Optimizer Statistics 16-7
16.3 Reporting on Past Statistics Gathering Operations 16-7
xv
19.2.2 Enabling Optimizer Features 19-8
19.2.3 Choosing an Optimizer Goal 19-9
19.2.4 Controlling Adaptive Optimization 19-10
19.3 Influencing the Optimizer with Hints 19-12
19.3.1 About Optimizer Hints 19-12
19.3.1.1 Types of Hints 19-14
19.3.1.2 Scope of Hints 19-15
19.3.1.3 Guidelines for Hints 19-15
19.3.2 Guidelines for Join Order Hints 19-16
xvi
Part VII Monitoring and Tracing SQL
xvii
23 Performing Application Tracing
23.1 Overview of End-to-End Application Tracing 23-1
23.1.1 Purpose of End-to-End Application Tracing 23-2
23.1.2 End-to-End Application Tracing in a Multitenant Environment 23-3
23.1.3 Tools for End-to-End Application Tracing 23-3
23.1.3.1 Overview of the SQL Trace Facility 23-4
23.1.3.2 Overview of TKPROF 23-5
23.2 Enabling Statistics Gathering for End-to-End Tracing 23-5
23.2.1 Enabling Statistics Gathering for a Client ID 23-5
23.2.2 Enabling Statistics Gathering for Services, Modules, and Actions 23-6
23.3 Enabling End-to-End Application Tracing 23-7
23.3.1 Enabling Tracing for a Client Identifier 23-8
23.3.2 Enabling Tracing for a Service, Module, and Action 23-8
23.3.3 Enabling Tracing for a Session 23-9
23.3.4 Enabling Tracing for the Instance or Database 23-10
23.4 Generating Output Files Using SQL Trace and TKPROF 23-11
23.4.1 Step 1: Setting Initialization Parameters for Trace File Management 23-12
23.4.2 Step 2: Enabling the SQL Trace Facility 23-14
23.4.3 Step 3: Generating Output Files with TKPROF 23-15
23.4.4 Step 4: Storing SQL Trace Facility Statistics 23-16
23.4.4.1 Generating the TKPROF Output SQL Script 23-17
23.4.4.2 Editing the TKPROF Output SQL Script 23-17
23.4.4.3 Querying the Output Table 23-17
23.5 Guidelines for Interpreting TKPROF Output 23-19
23.5.1 Guideline for Interpreting the Resolution of Statistics 23-19
23.5.2 Guideline for Recursive SQL Statements 23-19
23.5.3 Guideline for Deciding Which Statements to Tune 23-20
23.5.4 Guidelines for Avoiding Traps in TKPROF Interpretation 23-21
23.5.4.1 Guideline for Avoiding the Argument Trap 23-21
23.5.4.2 Guideline for Avoiding the Read Consistency Trap 23-21
23.5.4.3 Guideline for Avoiding the Schema Trap 23-22
23.5.4.4 Guideline for Avoiding the Time Trap 23-23
23.6.1 Application Tracing Utilities 23-23
23.6.1.1 TRCSESS 23-24
23.6.1.1.1 Purpose 23-24
23.6.1.1.2 Guidelines 23-24
23.6.1.1.3 Syntax 23-24
23.6.1.1.4 Options 23-25
23.6.1.1.5 Examples 23-25
23.6.1.2 TKPROF 23-26
xviii
23.6.1.2.1 Purpose 23-27
23.6.1.2.2 Guidelines 23-27
23.6.1.2.3 Syntax 23-27
23.6.1.2.4 Options 23-27
23.6.1.2.5 Output 23-29
23.6.1.2.6 Examples 23-33
23.7.1 Views for Application Tracing 23-39
23.7.1.1 Views Relevant for Trace Statistics 23-39
23.7.1.2 Views Related to Enabling Tracing 23-40
xix
25.1.3 SQL Tuning Advisor Operation 25-16
25.1.3.1 Automatic and On-Demand SQL Tuning 25-17
25.1.3.2 Local and Remote SQL Tuning 25-17
25.2 Managing the Automatic SQL Tuning Task 25-19
25.2.1 About the Automatic SQL Tuning Task 25-20
25.2.1.1 Purpose of Automatic SQL Tuning 25-20
25.2.1.2 Automatic SQL Tuning Concepts 25-20
25.2.1.3 Command-Line Interface to SQL Tuning Advisor 25-21
25.2.1.4 Basic Tasks for Automatic SQL Tuning 25-21
25.2.2 Enabling and Disabling the Automatic SQL Tuning Task 25-22
25.2.2.1 Enabling and Disabling the Automatic SQL Tuning Task Using
Cloud Control 25-23
25.2.2.2 Enabling and Disabling the Automatic SQL Tuning Task from the
Command Line 25-24
25.2.3 Configuring the Automatic SQL Tuning Task 25-25
25.2.3.1 Configuring the Automatic SQL Tuning Task Using Cloud
Control 25-26
25.2.3.2 Configuring the Automatic SQL Tuning Task Using the
Command Line 25-27
25.2.4 Viewing Automatic SQL Tuning Reports 25-28
25.2.4.1 Viewing Automatic SQL Tuning Reports Using the Command
Line 25-29
25.3 Running SQL Tuning Advisor On Demand 25-32
25.3.1 About On-Demand SQL Tuning 25-33
25.3.1.1 Purpose of On-Demand SQL Tuning 25-33
25.3.1.2 User Interfaces for On-Demand SQL Tuning 25-33
25.3.1.3 Basic Tasks in On-Demand SQL Tuning 25-35
25.3.2 Creating a SQL Tuning Task 25-36
25.3.3 Configuring a SQL Tuning Task 25-38
25.3.4 Executing a SQL Tuning Task 25-40
25.3.5 Monitoring a SQL Tuning Task 25-41
25.3.6 Displaying the Results of a SQL Tuning Task 25-42
xx
26.1.3 User Interfaces for SQL Access Advisor 26-8
26.1.3.1 Accessing the SQL Access Advisor: Initial Options Page Using
Cloud Control 26-9
26.1.3.2 Command-Line Interface to SQL Tuning Sets 26-10
26.2 Optimizing Access Paths with SQL Access Advisor: Basic Tasks 26-10
26.2.1 Creating a SQL Tuning Set as Input for SQL Access Advisor 26-12
26.2.2 Populating a SQL Tuning Set with a User-Defined Workload 26-13
26.2.3 Creating and Configuring a SQL Access Advisor Task 26-16
26.2.4 Executing a SQL Access Advisor Task 26-18
26.2.5 Viewing SQL Access Advisor Task Results 26-19
26.2.6 Generating and Executing a Task Script 26-23
26.3 Performing a SQL Access Advisor Quick Tune 26-25
26.4 Using SQL Access Advisor: Advanced Tasks 26-26
26.4.1 Evaluating Existing Access Structures 26-26
26.4.2 Updating SQL Access Advisor Task Attributes 26-27
26.4.3 Creating and Using SQL Access Advisor Task Templates 26-28
26.4.4 Terminating SQL Access Advisor Task Execution 26-30
26.4.4.1 Interrupting SQL Access Advisor Tasks 26-31
26.4.4.2 Canceling SQL Access Advisor Tasks 26-31
26.4.5 Deleting SQL Access Advisor Tasks 26-33
26.4.6 Marking SQL Access Advisor Recommendations 26-34
26.4.7 Modifying SQL Access Advisor Recommendations 26-35
26.5 SQL Access Advisor Examples 26-36
26.6 SQL Access Advisor Reference 26-36
26.6.1 Action Attributes in the DBA_ADVISOR_ACTIONS View 26-36
26.6.2 Categories for SQL Access Advisor Task Parameters 26-38
26.6.3 SQL Access Advisor Constants 26-39
xxi
27.2.2 Implementing a SQL Profile 27-9
27.3 Listing SQL Profiles 27-10
27.4 Altering a SQL Profile 27-11
27.5 Dropping a SQL Profile 27-12
27.6 Transporting a SQL Profile 27-13
xxii
29.2.2.1 About the SPM Evolve Advisor Task 29-11
29.2.2.2 Enabling and Disabling the SPM Evolve Advisor Task 29-12
29.2.2.3 Configuring the Automatic SPM Evolve Advisor Task 29-12
29.3 Displaying Plans in a SQL Plan Baseline 29-14
29.4 Loading SQL Plan Baselines 29-16
29.4.1 About Loading SQL Plan Baselines 29-16
29.4.2 Loading Plans from AWR 29-18
29.4.3 Loading Plans from the Shared SQL Area 29-20
29.4.4 Loading Plans from a SQL Tuning Set 29-22
29.4.5 Loading Plans from a Staging Table 29-25
29.5 Evolving SQL Plan Baselines Manually 29-27
29.5.1 About the DBMS_SPM Evolve Functions 29-28
29.5.2 Managing an Evolve Task 29-29
29.6 Dropping SQL Plan Baselines 29-39
29.7 Managing the SQL Management Base 29-40
29.7.1 About Managing the SMB 29-41
29.7.2 Changing the Disk Space Limit for the SMB 29-42
29.7.3 Changing the Plan Retention Policy in the SMB 29-43
xxiii
A.1.4 Guidelines for Writing SQL Statements That Use Indexes A-6
A.1.5 Guidelines for Writing SQL Statements That Avoid Using Indexes A-6
A.1.6 Guidelines for Re-Creating Indexes A-6
A.1.7 Guidelines for Compacting Indexes A-7
A.1.8 Guidelines for Using Nonunique Indexes to Enforce Uniqueness A-7
A.1.9 Guidelines for Using Enabled Novalidated Constraints A-8
A.2 Guidelines for Using Function-Based Indexes for Performance A-9
A.3 Guidelines for Using Partitioned Indexes for Performance A-10
A.4 Guidelines for Using Index-Organized Tables for Performance A-11
A.5 Guidelines for Using Bitmap Indexes for Performance A-11
A.6 Guidelines for Using Bitmap Join Indexes for Performance A-12
A.7 Guidelines for Using Domain Indexes for Performance A-12
A.8 Guidelines for Using Table Clusters A-13
A.9 Guidelines for Using Hash Clusters for Performance A-14
Glossary
Index
xxiv
Preface
This manual explains how to tune Oracle SQL.
This preface contains the following topics:
• Audience
• Documentation Accessibility
• Related Documents
• Conventions
Audience
This document is intended for database administrators and application developers who
perform the following tasks:
• Generating and interpreting SQL execution plans
• Managing optimizer statistics
• Influencing the optimizer through initialization parameters or SQL hints
• Controlling cursor sharing for SQL statements
• Monitoring SQL execution
• Performing application tracing
• Managing SQL tuning sets
• Using SQL Tuning Advisor or SQL Access Advisor
• Managing SQL profiles
• Managing SQL baselines
Documentation Accessibility
For information about Oracle's commitment to accessibility, visit the Oracle
Accessibility Program website at http://www.oracle.com/pls/topic/lookup?
ctx=acc&id=docacc.
xxv
Preface
Related Documents
This manual assumes that you are familiar with Oracle Database Concepts. The
following books are frequently referenced:
• Oracle Database Data Warehousing Guide
• Oracle Database VLDB and Partitioning Guide
• Oracle Database SQL Language Reference
• Oracle Database Reference
Many examples in this book use the sample schemas, which are installed by default
when you select the Basic Installation option with an Oracle Database. See Oracle
Database Sample Schemas for information on how these schemas were created and
how you can use them.
Conventions
The following text conventions are used in this document:
Convention Meaning
boldface Boldface type indicates graphical user interface elements associated
with an action, or terms defined in text or the glossary.
italic Italic type indicates book titles, emphasis, or placeholder variables for
which you supply particular values.
monospace Monospace type indicates commands within a paragraph, URLs, code
in examples, text that appears on the screen, or text that you enter.
xxvi
Changes in This Release for Oracle
Database SQL Tuning Guide
This preface describes the most important changes in Oracle Database SQL Tuning
Guide.
This preface contains the following topics:
• Changes in Oracle Database 12c Release 2 (12.2.0.1)
Oracle Database SQL Tuning Guide for Oracle Database 12c Release 2 (12.2.0.1)
has the following changes.
• Changes in Oracle Database 12c Release 1 (12.1.0.2)
Oracle Database SQL Tuning Guide for Oracle Database 12c Release 1 (12.1.0.2)
has the following changes.
• Changes in Oracle Database 12c Release 1 (12.1.0.1)
Oracle Database SQL Tuning Guide for Oracle Database 12c Release 1 (12.1)
has the following changes.
New Features
The following features are new in this release:
• Advisor enhancements
– Optimizer Statistics Advisor
Optimizer Statistics Advisor is built-in diagnostic software that analyzes the
quality of statistics and statistics-related tasks. The advisor task runs
automatically in the maintenance window, but you can also run it on demand.
You can then view the advisor report. If the advisor makes recommendations,
then in some cases you can run system-generated scripts to implement them.
See "Analyzing Statistics Using Optimizer Statistics Advisor".
xxvii
Changes in This Release for Oracle Database SQL Tuning Guide
xxviii
Changes in This Release for Oracle Database SQL Tuning Guide
xxix
Changes in This Release for Oracle Database SQL Tuning Guide
Desupported Features
The following features are desupported in Oracle Database 12c Release 2 (12.2.0.1).
• The OPTIMIZER_ADAPTIVE_FEATURES initialization parameter
See Also:
Oracle Database Upgrade Guide for a list of desupported features
Other Changes
This topic describes additional changes in the release.
• New Real-World Performance content
In this release, the book incorporates information provided by the Real-World
Performance group, including the following:
– "Improving Real-World Performance Through Cursor Sharing" explains how to
use bind variables and new features such as adaptive cursor sharing
New Features
The following features are new in this release.
• In-Memory aggregation
This optimization minimizes the join and GROUP BY processing required for each
row when joining a single large table to multiple small tables, as in a star schema.
VECTOR GROUP BY aggregation uses the infrastructure related to parallel query (PQ)
processing, and blends it with CPU-efficient algorithms to maximize the
performance and effectiveness of the initial aggregation performed before
redistributing fact data.
xxx
Changes in This Release for Oracle Database SQL Tuning Guide
New Features
The following features are new in this release.
• Adaptive SQL Plan Management (SPM)
The SPM Evolve Advisor is a task infrastructure that enables you to schedule an
evolve task, rerun an evolve task, and generate persistent reports. The new
automatic evolve task, SYS_AUTO_SPM_EVOLVE_TASK, runs in the default
maintenance window. This task ranks all unaccepted plans and runs the evolve
process for them. If the task finds a new plan that performs better than existing
plan, the task automatically accepts the plan. You can also run evolution tasks
manually using the DBMS_SPM package.
See "Managing the SPM Evolve Advisor Task".
• Adaptive query optimization
Adaptive query optimization is a set of capabilities that enable the optimizer to
make run-time adjustments to execution plans and discover additional information
that can lead to better statistics. The set of capabilities include:
– Adaptive query plans
An adaptive query plan has built-in options that enable the final plan for a
statement to differ from the default plan. During the first execution, before a
specific subplan becomes active, the optimizer makes a final decision about
xxxi
Changes in This Release for Oracle Database SQL Tuning Guide
which option to use. The optimizer bases its choice on observations made
during the execution up to this point. The ability of the optimizer to adapt plans
can improve query performance.
See "Adaptive Query Plans".
– Automatic reoptimization
When using automatic reoptimization, the optimizer monitors the initial
execution of a query. If the actual execution statistics vary significantly from
the original plan statistics, then the optimizer records the execution statistics
and uses them to choose a better plan the next time the statement executes.
The database uses information obtained during automatic reoptimization to
generate SQL plan directives automatically.
See "Automatic Reoptimization".
– SQL plan directives
In releases earlier than Oracle Database 12c, the database stored compilation
and execution statistics in the shared SQL area, which is nonpersistent.
Starting in this release, the database can use a SQL plan directive, which is
additional information and instructions that the optimizer can use to generate a
more optimal plan. The database stores SQL plan directives persistently in the
SYSAUX tablespace. When generating an execution plan, the optimizer can use
SQL plan directives to obtain more information about the objects accessed in
the plan.
See "SQL Plan Directives".
– Dynamic statistics enhancements
In releases earlier than Oracle Database 12c, Oracle Database only used
dynamic statistics (previously called dynamic sampling) when one or more of
the tables in a query did not have optimizer statistics. Starting in this release,
the optimizer automatically decides whether dynamic statistics are useful and
which dynamic statistics level to use for all SQL statements. Dynamic statistics
gathers are persistent and usable by other queries.
See "Supplemental Dynamic Statistics".
• New types of histograms
This release introduces top frequency and hybrid histograms. If a column contains
more than 254 distinct values, and if the top 254 most frequent values occupy
more than 99% of the data, then the database creates a top frequency histogram
using the top 254 most frequent values. By ignoring the nonpopular values, which
are statistically insignificant, the database can produce a better quality histogram
for highly popular values. A hybrid histogram is an enhanced height-based
histogram that stores the exact frequency of each endpoint in the sample, and
ensures that a value is never stored in multiple buckets.
Also, regular frequency histograms have been enhanced. The optimizer computes
frequency histograms during NDV computation based on a full scan of the data
rather than a small sample (when AUTO_SAMPLING is used). The enhanced
frequency histograms ensure that even highly infrequent values are properly
represented with accurate bucket counts within a histogram.
See "Histograms ".
• Monitoring database operations
xxxii
Changes in This Release for Oracle Database SQL Tuning Guide
xxxiii
Changes in This Release for Oracle Database SQL Tuning Guide
Deprecated Features
The following features are deprecated in this release, and may be desupported in a
future release.
• Stored outlines
See "Managing SQL Plan Baselines" for information about alternatives.
• The SIMILAR value for the CURSOR_SHARING initialization parameter
This value is deprecated. Use FORCE instead.
See "Do Not Use CURSOR_SHARING = FORCE as a Permanent Fix".
xxxiv
Changes in This Release for Oracle Database SQL Tuning Guide
Desupported Features
Some features previously described in this document are desupported in Oracle
Database 12c.
See Oracle Database Upgrade Guide for a list of desupported features.
Other Changes
The following are additional changes in the release.
• New tuning books
The Oracle Database 11g Oracle Database Performance Tuning Guide has been
divided into two books for Oracle Database 12c:
– Oracle Database Performance Tuning Guide, which contains only topics that
pertain to tuning the database
– Oracle Database SQL Tuning Guide, which contains only topics that pertain to
tuning SQL
xxxv
Part I
SQL Performance Fundamentals
SQL tuning is improving SQL statement performance to meet specific, measurable,
and achievable goals.
This part contains the following chapters:
• Introduction to SQL Tuning
SQL tuning is the attempt to diagnose and repair SQL statements that fail to meet
a performance standard.
• SQL Performance Methodology
This chapter describes the recommended methodology for SQL tuning.
1
Introduction to SQL Tuning
SQL tuning is the attempt to diagnose and repair SQL statements that fail to meet a
performance standard.
This chapter contains the following topics:
• About SQL Tuning
SQL tuning is the iterative process of improving SQL statement performance to
meet specific, measurable, and achievable goals.
• Purpose of SQL Tuning
A SQL statement becomes a problem when it fails to perform according to a
predetermined and measurable standard.
• Prerequisites for SQL Tuning
SQL performance tuning requires a foundation of database knowledge.
• Tasks and Tools for SQL Tuning
After you have identified the goal for a tuning session, for example, reducing user
response time from three minutes to less than a second, the problem becomes
how to accomplish this goal.
See Also:
1-1
Chapter 1
Prerequisites for SQL Tuning
• Improve throughput, which means using the least amount of resources necessary
to process all rows accessed by a statement
For a response time problem, consider an online book seller application that hangs for
three minutes after a customer updates the shopping cart. Contrast with a three-
minute parallel query in a data warehouse that consumes all of the database host
CPU, preventing other queries from running. In each case, the user response time is
three minutes, but the cause of the problem is different, and so is the tuning goal.
1-2
Chapter 1
Tasks and Tools for SQL Tuning
1-3
Chapter 1
Tasks and Tools for SQL Tuning
query predicate with low selectivity may use a full table scan on a large table
instead of an index.
You can compare the execution plan of an optimally performing SQL
statement to the plan of the statement when it performs suboptimally. This
comparison, along with information such as changes in data volumes, can
help identify causes of performance degradation.
• Missing SQL access structures
Absence of SQL access structures, such as indexes and materialized views, is
a typical reason for suboptimal SQL performance. The optimal set of access
structures can improve SQL performance by orders of magnitude.
• Stale optimizer statistics
Statistics gathered by DBMS_STATS can become stale when the statistics
maintenance operations, either automatic or manual, cannot keep up with the
changes to the table data caused by DML. Because stale statistics on a table
do not accurately reflect the table data, the optimizer can make decisions
based on faulty information and generate suboptimal execution plans.
• Hardware problems
Suboptimal performance might be connected with memory, I/O, and CPU
problems.
4. Defining the scope of the problem
The scope of the solution must match the scope of the problem. Consider a
problem at the database level and a problem at the statement level. For example,
the shared pool is too small, which causes cursors to age out quickly, which in turn
causes many hard parses. Using an initialization parameter to increase the shared
pool size fixes the problem at the database level and improves performance for all
sessions. However, if a single SQL statement is not using a helpful index, then
changing the optimizer initialization parameters for the entire database could harm
overall performance. If a single SQL statement has a problem, then an
appropriately scoped solution addresses just this problem with this statement.
5. Implementing corrective actions for suboptimally performing SQL statements
These actions vary depending on circumstances. For example, you might rewrite a
SQL statement to be more efficient, avoiding unnecessary hard parsing by
rewriting the statement to use bind variables. You might also use equijoins,
remove functions from WHERE clauses, and break a complex SQL statement into
multiple simple statements.
In some cases, you improve SQL performance not by rewriting the statement, but
by restructuring schema objects. For example, you might index a new access
path, or reorder columns in a concatenated index. You might also partition a table,
introduce derived values, or even change the database design.
6. Preventing SQL performance regressions
To ensure optimal SQL performance, verify that execution plans continue to
provide optimal performance, and choose better plans if they come available. You
can achieve these goals using optimizer statistics, SQL profiles, and SQL plan
baselines.
1-4
Chapter 1
Tasks and Tools for SQL Tuning
See Also:
1-5
Chapter 1
Tasks and Tools for SQL Tuning
See Also:
See Also:
1-6
Chapter 1
Tasks and Tools for SQL Tuning
See Also:
See Also:
1-7
Chapter 1
Tasks and Tools for SQL Tuning
See Also:
See Also:
Oracle Database Testing Guide
1-8
Chapter 1
Tasks and Tools for SQL Tuning
1-9
Chapter 1
Tasks and Tools for SQL Tuning
See Also:
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_MONITOR and DBMS_SESSION
1-10
Chapter 1
Tasks and Tools for SQL Tuning
selective for certain queries. In this case, you may use hints to instruct the optimizer to
use a better execution plan, as in the following example:
See Also:
Oracle recommends Cloud Control as the best interface for database administration
and tuning. In cases where the command-line interface better illustrates a particular
concept or task, this manual uses command-line examples. However, in these cases
the tuning tasks include a reference to the principal Cloud Control page associated
with the task.
1-11
2
SQL Performance Methodology
This chapter describes the recommended methodology for SQL tuning.
Note:
This book assumes that you have learned the Oracle Database performance
methodology described in Oracle Database 2 Day + Performance Tuning
Guide.
2-1
Chapter 2
Guidelines for Designing Your Application
2-2
Chapter 2
Guidelines for Deploying Your Application
SELECT *
FROM employees
WHERE last_name LIKE 'KING';
SELECT *
FROM employees
WHERE last_name LIKE :1;
The following example shows the results of some tests on a simple OLTP
application:
2-3
Chapter 2
Guidelines for Deploying Your Application
All testing must be done with fully populated tables. The test database should
contain data representative of the production system in terms of data volume and
cardinality between tables. All the production indexes should be built and the
schema statistics should be populated correctly.
• Use the correct optimizer mode.
Perform all testing with the optimizer mode that you plan to use in production.
• Test a single user performance.
Test a single user on an idle or lightly-used database for acceptable performance.
If a single user cannot achieve acceptable performance under ideal conditions,
then multiple users cannot achieve acceptable performance under real conditions.
• Obtain and document plans for all SQL statements.
Obtain an execution plan for each SQL statement. Use this process to verify that
the optimizer is obtaining an optimal execution plan, and that the relative cost of
the SQL statement is understood in terms of CPU time and physical I/Os. This
process assists in identifying the heavy use transactions that require the most
tuning and performance work in the future.
• Attempt multiuser testing.
This process is difficult to perform accurately, because user workload and profiles
might not be fully quantified. However, transactions performing DML statements
should be tested to ensure that there are no locking conflicts or serialization
problems.
• Test with the correct hardware configuration.
Test with a configuration as close to the production system as possible. Using a
realistic system is particularly important for network latencies, I/O subsystem
bandwidth, and processor type and speed. Failing to use this approach may result
in an incorrect analysis of potential performance problems.
• Measure steady state performance.
When benchmarking, it is important to measure the performance under steady
state conditions. Each benchmark run should have a ramp-up phase, where users
are connected to the application and gradually start performing work on the
application. This process allows for frequently cached data to be initialized into the
cache and single execution operations—such as parsing—to be completed before
the steady state condition. Likewise, after a benchmark run, a ramp-down period is
useful so that the system frees resources, and users cease work and disconnect.
2-4
Chapter 2
Guidelines for Deploying Your Application
It is difficult to recommend one approach over the other, because each technique has
associated risks that could lead to system outages as the transition takes place.
Certainly, the Trickle approach allows profiling of real users as they are introduced to
the new application, and allows the system to be reconfigured while only affecting the
migrated users. This approach affects the work of the early adopters, but limits the
load on support services. Thus, unscheduled outages only affect a small percentage of
the user population.
The decision on how to roll out a new application is specific to each business. Any
adopted approach has its own unique pressures and stresses. The more testing and
knowledge that you derive from the testing process, the more you realize what is best
for the rollout.
2-5
Part II
Query Optimizer Fundamentals
To tune Oracle SQL, you must understand the query optimizer. The optimizer is built-
in software that determines the most efficient method for a statement to access data.
This part contains the following chapters:
• SQL Processing
This chapter explains how database processes DDL statements to create objects,
DML to modify data, and queries to retrieve data.
• Query Optimizer Concepts
This chapter describes the most important concepts relating to the query
optimizer, including its principal components.
• Query Transformations
The optimizer employs many query transformation techniques. This chapter
describes some of the most important.
3
SQL Processing
This chapter explains how database processes DDL statements to create objects,
DML to modify data, and queries to retrieve data.
This chapter contains the following topics:
• About SQL Processing
SQL processing is the parsing, optimization, row source generation, and
execution of a SQL statement.
• How Oracle Database Processes DML
Most DML statements have a query component. In a query, execution of a cursor
places the results of the query into a set of rows called the result set.
• How Oracle Database Processes DDL
Oracle Database processes DDL differently from DML.
3-1
Chapter 3
About SQL Processing
SQL Statement
Parsing
Syntax
Check
Semantic
Check
Hard Parse
Generation of
multiple Optimization
execution plans
Execution
3-2
Chapter 3
About SQL Processing
• Syntax Check
Oracle Database must check each SQL statement for syntactic validity.
• Semantic Check
The semantics of a statement are its meaning. A semantic check determines
whether a statement is meaningful, for example, whether the objects and columns
in the statement exist.
• Shared Pool Check
During the parse, the database performs a shared pool check to determine
whether it can skip resource-intensive steps of statement processing.
See Also:
Oracle Database Concepts to learn about deadlocks
3-3
Chapter 3
About SQL Processing
Note:
The database always performs a hard parse of DDL.
3-4
Chapter 3
About SQL Processing
During the hard parse, the database accesses the library cache and data
dictionary cache numerous times to check the data dictionary. When the database
accesses these areas, it uses a serialization device called a latch on required
objects so that their definition does not change. Latch contention increases
statement execution time and decreases concurrency.
• Soft parse
A soft parse is any parse that is not a hard parse. If the submitted statement is the
same as a reusable SQL statement in the shared pool, then Oracle Database
reuses the existing code. This reuse of code is also called a library cache hit.
Soft parses can vary in how much work they perform. For example, configuring the
session shared SQL area can sometimes reduce the amount of latching in the soft
parses, making them "softer."
In general, a soft parse is preferable to a hard parse because the database skips
the optimization and row source generation steps, proceeding straight to
execution.
The following graphic is a simplified representation of a shared pool check of an
UPDATE statement in a dedicated server architecture.
Shared Pool
Library Cache
Shared SQL Area Private
3667723989 SQL Area
3967354608
2190280494
User
Private SQL Area
If a check determines that a statement in the shared pool has the same hash value,
then the database performs semantic and environment checks to determine whether
the statements have the same meaning. Identical syntax is not sufficient. For example,
suppose two different users log in to the database and issue the following SQL
statements:
3-5
Chapter 3
About SQL Processing
The SELECT statements for the two users are syntactically identical, but two separate
schema objects are named my_table. This semantic difference means that the second
statement cannot reuse the code for the first statement.
Even if two statements are semantically identical, an environmental difference can
force a hard parse. In this context, the optimizer environment is the totality of session
settings that can affect execution plan generation, such as the work area size or
optimizer settings (for example, the optimizer mode). Consider the following series of
SQL statements executed by a single user:
In the preceding example, the same SELECT statement is executed in three different
optimizer environments. Consequently, the database creates three separate shared
SQL areas for these statements and forces a hard parse of each statement.
See Also:
• Oracle Database Concepts to learn about private SQL areas and shared
SQL areas
• Oracle Database Performance Tuning Guide to learn how to configure
the shared pool
• Oracle Database Concepts to learn about latches
3-6
Chapter 3
About SQL Processing
Execution Plan
----------------------------------------------------------
Plan hash value: 975837011
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 3 | 189 | 7(15)|
00:00:01 |
|*1 | HASH JOIN | | 3 | 189 | 7(15)|
00:00:01 |
|*2 | HASH JOIN | | 3 | 141 | 5(20)|
00:00:01 |
| 3 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES | 3 | 60 | 2 (0)|
00:00:01 |
3-7
Chapter 3
About SQL Processing
1 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
2 - access("E"."JOB_ID"="J"."JOB_ID")
4 - access("E"."LAST_NAME" LIKE 'A%')
filter("E"."LAST_NAME" LIKE 'A%')
3-8
Chapter 3
About SQL Processing
1
HASH JOIN
2 6
HASH JOIN TABLE ACCESS
FULL
departments
3 5
TABLE ACCESS TABLE ACCESS
BY INDEX ROWID FULL
employees jobs
4
INDEX RANGE
SCAN
emp_name_ix
In Figure 3-3, each node of the tree acts as a row source, which means that each step
of the execution plan in Example 3-1 either retrieves rows from the database or
accepts rows from one or more row sources as input. The SQL engine executes each
row source as follows:
• Steps indicated by the black boxes physically retrieve data from an object in the
database. These steps are the access paths, or techniques for retrieving data from
the database.
– Step 6 uses a full table scan to retrieve all rows from the departments table.
– Step 5 uses a full table scan to retrieve all rows from the jobs table.
– Step 4 scans the emp_name_ix index in order, looking for each key that begins
with the letter A and retrieving the corresponding rowid. For example, the rowid
corresponding to Atkinson is AAAPzRAAFAAAABSAAe.
– Step 3 retrieves from the employees table the rows whose rowids were
returned by Step 4. For example, the database uses rowid
AAAPzRAAFAAAABSAAe to retrieve the row for Atkinson.
• Steps indicated by the clear boxes operate on row sources.
3-9
Chapter 3
How Oracle Database Processes DML
– Step 2 performs a hash join, accepting row sources from Steps 3 and 5,
joining each row from the Step 5 row source to its corresponding row in Step
3, and returning the resulting rows to Step 1.
For example, the row for employee Atkinson is associated with the job name
Stock Clerk.
– Step 1 performs another hash join, accepting row sources from Steps 2 and 6,
joining each row from the Step 6 source to its corresponding row in Step 2,
and returning the result to the client.
For example, the row for employee Atkinson is associated with the
department named Shipping.
In some execution plans the steps are iterative and in others sequential. The hash join
shown in Example 3-1 is sequential. The database completes the steps in their entirety
based on the join order. The database starts with the index range scan of
emp_name_ix. Using the rowids that it retrieves from the index, the database reads the
matching rows in the employees table, and then scans the jobs table. After it retrieves
the rows from the jobs table, the database performs the hash join.
During execution, the database reads the data from disk into memory if the data is not
in memory. The database also takes out any locks and latches necessary to ensure
data integrity and logs any changes made during the SQL execution. The final stage of
processing a SQL statement is closing the cursor.
3-10
Chapter 3
How Oracle Database Processes DDL
See Also:
Oracle Database Concepts to learn about multiversion read consistency
Typically, the database would run dozens of recursive statements to execute the
preceding statement. The recursive SQL would perform actions such as the following:
• Issue a COMMIT before executing the CREATE TABLE statement
• Verify that user privileges are sufficient to create the table
• Determine which tablespace the table should reside in
• Ensure that the tablespace quota has not been exceeded
3-11
Chapter 3
How Oracle Database Processes DDL
See Also:
Oracle Database Development Guide to learn about processing DDL,
transaction control, and other types of statements
3-12
4
Query Optimizer Concepts
This chapter describes the most important concepts relating to the query optimizer,
including its principal components.
This chapter contains the following topics:
• Introduction to the Query Optimizer
The query optimizer (called simply the optimizer) is built-in database software
that determines the most efficient method for a SQL statement to access
requested data.
• About Optimizer Components
The optimizer contains three components: the transformer, estimator, and plan
generator.
• About Automatic Tuning Optimizer
The optimizer performs different operations depending on how it is invoked.
• About Adaptive Query Optimization
In Oracle Database, adaptive query optimization enables the optimizer to make
run-time adjustments to execution plans and discover additional information that
can lead to better statistics.
• About Approximate Query Processing
Approximate query processing is a set of optimization techniques that speed
analytic queries by calculating results within an acceptable range of error.
• About SQL Plan Management
SQL plan management enables the optimizer to automatically manage execution
plans, ensuring that the database uses only known or verified plans.
• About the Expression Statistics Store (ESS)
The Expression Statistics Store (ESS) is a repository maintained by the
optimizer to store statistics about expression evaluation.
4-1
Chapter 4
Introduction to the Query Optimizer
Note:
The optimizer may not make the same decisions from one version of Oracle
Database to the next. In recent versions, the optimizer might make different
decision because better information is available and more optimizer
transformations are possible.
Related Topics
• Cost
The optimizer cost model accounts for the machine resources that a query is
predicted to use.
4-2
Chapter 4
Introduction to the Query Optimizer
GB Plan GB Plan
1 2
NL HJ
NL HJ
Generates Multiple
Plans and
Compares Them
Statistics
1 0 1 1 0 0 1 0 0
4-3
Chapter 4
Introduction to the Query Optimizer
Each SELECT block in the original SQL statement is represented internally by a query
block. A query block can be a top-level statement, subquery, or unmerged view.
Example 4-1 Query Blocks
The following SQL statement consists of two query blocks. The subquery in
parentheses is the inner query block. The outer query block, which is the rest of the
SQL statement, retrieves names of employees in the departments whose IDs were
supplied by the subquery. The query form determines how query blocks are
interrelated.
See Also:
• "View Merging"
• Oracle Database Concepts for an overview of SQL processing
4-4
Chapter 4
About Optimizer Components
Parsed Query
(from Parser)
Query
Transformer
Transformed query
statistics Data
Estimator Dictionary
Query + estimates
Plan
Generator
Query Plan
(to Row Source Generator)
A set of query blocks represents a parsed query, which is the input to the optimizer.
The following table describes the optimizer operations.
4-5
Chapter 4
About Optimizer Components
4-6
Chapter 4
About Optimizer Components
SELECT *
FROM sales
WHERE promo_id=33
OR prod_id=136;
Query Transformer
SELECT *
FROM sales
WHERE prod_id=136
UNION ALL
SELECT *
FROM sales
WHERE promo_id=33
AND LNNVL(prod_id=136);
Related Topics
• Query Transformations
The optimizer employs many query transformation techniques. This chapter
describes some of the most important.
4.2.2 Estimator
The estimator is the component of the optimizer that determines the overall cost of a
given execution plan.
The estimator uses three different measures to determine cost:
• Selectivity
The percentage of rows in the row set that the query selects, with 0 meaning no
rows and 1 meaning all rows. Selectivity is tied to a query predicate, such as WHERE
last_name LIKE 'A%', or a combination of predicates. A predicate becomes more
selective as the selectivity value approaches 0 and less selective (or more
unselective) as the value approaches 1.
Note:
Selectivity is an internal calculation that is not visible in the execution
plans.
• Cardinality
The cardinality is the number of rows returned by each operation in an execution
plan. This input, which is crucial to obtaining an optimal plan, is common to all cost
functions. The estimator can derive cardinality from the table statistics collected by
4-7
Chapter 4
About Optimizer Components
DBMS_STATS, or derive it after accounting for effects from predicates (filter, join, and
so on), DISTINCT or GROUP BY operations, and so on. The Rows column in an
execution plan shows the estimated cardinality.
• Cost
This measure represents units of work or resource used. The query optimizer uses
disk I/O, CPU usage, and memory usage as units of work.
As shown in the following graphic, if statistics are available, then the estimator uses
them to compute the measures. The statistics improve the degree of accuracy of the
measures.
Cardinality
Selectivity Cost
GB Plan
Estimator
HJ
HJ Total Cost
Statistics
1 0 1 0 0
0 0 0 1 1
0 1 1 0 1
For the query shown in Example 4-1, the estimator uses selectivity, estimated
cardinality (a total return of 10 rows), and cost measures to produce its total cost
estimate of 3:
---------------------------------------------------------------------------
-----
| Id| Operation |Name |Rows|Bytes|Cost %CPU|
Time|
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 10| 250| 3 (0)|
00:00:01|
| 1 | NESTED LOOPS | | | |
| |
| 2 | NESTED LOOPS | | 10| 250| 3 (0)|
00:00:01|
|*3 | TABLE ACCESS FULL |DEPARTMENTS | 1| 7| 2 (0)|
00:00:01|
|*4 | INDEX RANGE SCAN |EMP_DEPARTMENT_IX| 10| | 0 (0)|
00:00:01|
| 5 | TABLE ACCESS BY INDEX ROWID|EMPLOYEES | 10| 180| 1 (0)|
00:00:01|
---------------------------------------------------------------------------
-----
4-8
Chapter 4
About Optimizer Components
• Selectivity
The selectivity represents a fraction of rows from a row set.
• Cardinality
The cardinality is the number of rows returned by each operation in an execution
plan.
• Cost
The optimizer cost model accounts for the machine resources that a query is
predicted to use.
4.2.2.1 Selectivity
The selectivity represents a fraction of rows from a row set.
The row set can be a base table, a view, or the result of a join. The selectivity is tied to
a query predicate, such as last_name = 'Smith', or a combination of predicates, such
as last_name = 'Smith' AND job_id = 'SH_CLERK'.
Note:
Selectivity is an internal calculation that is not visible in execution plans.
A predicate filters a specific number of rows from a row set. Thus, the selectivity of a
predicate indicates how many rows pass the predicate test. Selectivity ranges from 0.0
to 1.0. A selectivity of 0.0 means that no rows are selected from a row set, whereas a
selectivity of 1.0 means that all rows are selected. A predicate becomes more
selective as the value approaches 0.0 and less selective (or more unselective) as the
value approaches 1.0.
The optimizer estimates selectivity depending on whether statistics are available:
• Statistics not available
Depending on the value of the OPTIMIZER_DYNAMIC_SAMPLING initialization
parameter, the optimizer either uses dynamic statistics or an internal default value.
The database uses different internal defaults depending on the predicate type. For
example, the internal default for an equality predicate (last_name = 'Smith') is
lower than for a range predicate (last_name > 'Smith') because an equality
predicate is expected to return a smaller fraction of rows.
• Statistics available
When statistics are available, the estimator uses them to estimate selectivity.
Assume there are 150 distinct employee last names. For an equality predicate
last_name = 'Smith', selectivity is the reciprocal of the number n of distinct
values of last_name, which in this example is .006 because the query selects rows
that contain 1 out of 150 distinct values.
If a histogram exists on the last_name column, then the estimator uses the
histogram instead of the number of distinct values. The histogram captures the
distribution of different values in a column, so it yields better selectivity estimates,
especially for columns that have data skew.
4-9
Chapter 4
About Optimizer Components
See Also:
• "Histograms "
• Oracle Database Reference to learn more about
OPTIMIZER_DYNAMIC_SAMPLING
4.2.2.2 Cardinality
The cardinality is the number of rows returned by each operation in an execution
plan.
For example, if the optimizer estimate for the number of rows returned by a full table
scan is 100, then the cardinality estimate for this operation is 100. The cardinality
estimate appears in the Rows column of the execution plan.
The optimizer determines the cardinality for each operation based on a complex set of
formulas that use both table and column level statistics, or dynamic statistics, as input.
The optimizer uses one of the simplest formulas when a single equality predicate
appears in a single-table query, with no histogram. In this case, the optimizer assumes
a uniform distribution and calculates the cardinality for the query by dividing the total
number of rows in the table by the number of distinct values in the column used in the
WHERE clause predicate.
The employees table contains 107 rows. The current database statistics indicate that
the number of distinct values in the salary column is 58. Therefore, the optimizer
estimates the cardinality of the result set as 2, using the formula 107/58=1.84.
4.2.2.3 Cost
The optimizer cost model accounts for the machine resources that a query is
predicted to use.
The cost is an internal numeric measure that represents the estimated resource usage
for a plan. The cost is specific to a query in an optimizer environment. To estimate
cost, the optimizer considers factors such as the following:
• System resources, which includes estimated I/O, CPU, and memory
• Estimated number of rows returned (cardinality)
4-10
Chapter 4
About Optimizer Components
Note:
The cost is an internal measure that the optimizer uses to compare different
plans for the same query. You cannot tune or change cost.
The execution time is a function of the cost, but cost does not equate directly to time.
For example, if the plan for query A has a lower cost than the plan for query B, then
the following outcomes are possible:
• A executes faster than B.
• A executes slower than B.
• A executes in the same amount of time as B.
Therefore, you cannot compare the costs of different queries with one another. Also,
you cannot compare the costs of semantically equivalent queries that use different
optimizer modes.
4-11
Chapter 4
About Optimizer Components
Optimizer
Transformer
Access Path
Index
Full Table Scan
Hash Join
departments 0, employees 1
The following snippet from an optimizer trace file shows some computations that the
optimizer performs:
GENERAL PLANS
***************************************
Considering cardinality-based initial join order.
Permutations for Starting Table :0
Join order[1]: DEPARTMENTS[D]#0 EMPLOYEES[E]#1
***************
Now joining: EMPLOYEES[E]#1
***************
NL Join
Outer table: Card: 27.00 Cost: 2.01 Resp: 2.01 Degree: 1 Bytes: 16
Access path analysis for EMPLOYEES
. . .
Best NL cost: 13.17
. . .
SM Join
SM cost: 6.08
resc: 6.08 resc_io: 4.00 resc_cpu: 2501688
resp: 6.08 resp_io: 4.00 resp_cpu: 2501688
. . .
SM Join (with index on outer)
Access Path: index (FullScan)
. . .
HA Join
4-12
Chapter 4
About Automatic Tuning Optimizer
HA cost: 4.57
resc: 4.57 resc_io: 4.00 resc_cpu: 678154
resp: 4.57 resp_io: 4.00 resp_cpu: 678154
Best:: JoinMethod: Hash
Cost: 4.57 Degree: 1 Resp: 4.57 Card: 106.00 Bytes: 27
. . .
***********************
Join order[2]: EMPLOYEES[E]#1 DEPARTMENTS[D]#0
. . .
***************
Now joining: DEPARTMENTS[D]#0
***************
. . .
HA Join
HA cost: 4.58
resc: 4.58 resc_io: 4.00 resc_cpu: 690054
resp: 4.58 resp_io: 4.00 resp_cpu: 690054
Join order aborted: cost > best plan cost
***********************
The trace file shows the optimizer first trying the departments table as the outer table
in the join. The optimizer calculates the cost for three different join methods: nested
loops join (NL), sort merge (SM), and hash join (HA). The optimizer picks the hash join
as the most efficient method:
The optimizer then tries a different join order, using employees as the outer table. This
join order costs more than the previous join order, so it is abandoned.
The optimizer uses an internal cutoff to reduce the number of plans it tries when
finding the lowest-cost plan. The cutoff is based on the cost of the current best plan. If
the current best cost is large, then the optimizer explores alternative plans to find a
lower cost plan. If the current best cost is small, then the optimizer ends the search
swiftly because further cost improvement is not significant.
4-13
Chapter 4
About Adaptive Query Optimization
analysis to further improve the plan produced in normal mode. The optimizer
output is not an execution plan, but a series of actions, along with their rationale
and expected benefit for producing a significantly better plan.
See Also:
Adaptive Query
Optimization
Adaptive Adaptive
Plans Statistics
4-14
Chapter 4
About Adaptive Query Optimization
Dynamic Plans
To change plans at runtime, adaptive query plans use a dynamic plan, which is
represented as a set of subplan groups. A subplan group is a set of subplans. A
subplan is a portion of a plan that the optimizer can switch to as an alternative at run
time. For example, a nested loops join could switch to a hash join during execution.
The optimizer decides which subplan to use at run time. When notified of a new
statistic value relevant to a subplan group, the coordinator dispatches it to the handler
function for this subgroup.
4-15
Chapter 4
About Adaptive Query Optimization
Dynamic Plan
GB GB
Subplan Subplan
HJ HJ
HJ HJ
4-16
Chapter 4
About Adaptive Query Optimization
3. Based on the statistics gathered by the collector, the optimizer chooses a subplan.
The dynamic plan coordinator decides which subplan to use at runtime for all such
subplan groups. When notified of a new statistic value relevant to a subplan group,
the coordinator dispatches it to the handler function for this subgroup.
4. The collector stops collecting statistics and buffering rows, permitting rows to pass
through instead.
5. The database stores the adaptive plan in the child cursor, so that the next
execution of the statement can use it.
On subsequent executions of the child cursor, the optimizer continues to use the same
adaptive plan unless one of the following conditions is true, in which case it picks a
new plan for the current execution:
• The current plan ages out of the shared pool.
• A different optimizer feature (for example, adaptive cursor sharing or statistics
feedback) invalidates the current plan.
This section contains the following topics:
• Adaptive Query Plans: Join Method Example
This example shows how the optimizer can choose a different plan based on
information collected at runtime.
• Adaptive Query Plans: Parallel Distribution Methods
Typically, parallel execution requires data redistribution to perform operations such
as parallel sorts, aggregations, and joins.
• Adaptive Query Plans: Bitmap Index Pruning
Adaptive plans prune indexes that do not significantly reduce the number of
matched rows.
Related Topics
• Reading Adaptive Query Plans
The adaptive optimizer is a feature of the optimizer that enables it to adapt plans
based on run-time statistics. All adaptive mechanisms can execute a final plan for
a statement that differs from the default plan.
• Controlling Adaptive Optimization
In Oracle Database, adaptive query optimization is the process by which the
optimizer adapts an execution plan based on statistics collected at run time.
SELECT product_name
FROM order_items o, prod_info p
WHERE o.unit_price = 15
AND quantity > 1
AND p.product_id = o.product_id
4-17
Chapter 4
About Adaptive Query Optimization
An adaptive query plan for this statement shows two possible plans, one with a nested
loops join and the other with a hash join:
-----------------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes|Cost (%CPU)|Time|
-----------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | |7(100)| |
| * 1| HASH JOIN | |4| 128 | 7 (0)|00:00:01|
|- 2| NESTED LOOPS | |4| 128 | 7 (0)|00:00:01|
|- 3| NESTED LOOPS | |4| 128 | 7 (0)|00:00:01|
|- 4| STATISTICS COLLECTOR | | | | | |
| * 5| TABLE ACCESS FULL | ORDER_ITEMS |4| 48 | 3 (0)|00:00:01|
|-* 6| INDEX UNIQUE SCAN | PROD_INFO_PK |1| | 0 (0)| |
|- 7| TABLE ACCESS BY INDEX ROWID| PROD_INFO |1| 20 | 1 (0)|00:00:01|
| 8| TABLE ACCESS FULL | PROD_INFO |1| 20 | 1 (0)|00:00:01|
-----------------------------------------------------------------------------
1 - access("P"."PRODUCT_ID"="O"."PRODUCT_ID")
5 - filter(("O"."UNIT_PRICE"=15 AND "QUANTITY">1))
6 - access("P"."PRODUCT_ID"="O"."PRODUCT_ID")
Note
-----
- this is an adaptive plan (rows marked '-' are inactive)
A nested loops join is preferable if the database can avoid scanning a significant
portion of prod_info because its rows are filtered by the join predicate. If few rows are
filtered, however, then scanning the right table in a hash join is preferable.
The following graphic shows the adaptive process. For the query in the preceding
example, the adaptive portion of the default plan contains two subplans, each of which
uses a different join method. The optimizer automatically determines when each join
method is optimal, depending on the cardinality of the left side of the join.
The statistics collector buffers enough rows coming from the order_items table to
determine which join method to use. If the row count is below the threshold determined
by the optimizer, then the optimizer chooses the nested loops join; otherwise, the
optimizer chooses the hash join. In this case, the row count coming from the
order_items table is above the threshold, so the optimizer chooses a hash join for the
final plan, and disables buffering.
4-18
Chapter 4
About Adaptive Query Optimization
Nested Hash
Loops Join
Statistics
Collector
The optimizer disables the statistics collector after making the decision,
and lets the rows pass through.
Nested Hash
Loops Join
Statistics
Collector
The Note section of the execution plan indicates whether the plan is adaptive, and
which rows in the plan are inactive.
See Also:
4-19
Chapter 4
About Adaptive Query Optimization
Broadcast Distribution
The following graphic depicts a hybrid hash join between the departments and
employees tables, with a query coordinator directing 8 parallel server processes: P5-
P8 are producers, whereas P1-P4 are consumers. Each producer has its own
consumer.
4-20
Chapter 4
About Adaptive Query Optimization
Query
Coordinator
P1 P2 P3 P4
departments P5 employees
2
P6
3
P7
4
P8
The database inserts a statistics collector in front of each producer process scanning
the departments table. The query coordinator aggregates the collected statistics. The
distribution method is based on the run-time statistics. In Figure 4-9, the number of
rows is below the threshold (8), which is twice the DOP (4), so the optimizer chooses a
broadcast technique for the departments table.
See Also:
Oracle Database VLDB and Partitioning Guide to learn more about parallel
data redistribution techniques
4-21
Chapter 4
About Adaptive Query Optimization
The following sample execution plan shows that the query generated no rows for the
bitmap node in Step 12 and Step 17. The adaptive optimizer determined that filtering
rows by using the CAR_MODEL_IDX and CAR_MAKE_IDX indexes was inefficient. The
query did not use the steps in the plan that begin with a dash (-).
-----------------------------------------------------------
| Id | Operation | Name |
-----------------------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | SORT GROUP BY NOSORT | |
| 2 | HASH JOIN | |
| 3 | VIEW | VW_ST_5497B905 |
| 4 | NESTED LOOPS | |
| 5 | BITMAP CONVERSION TO ROWIDS | |
| 6 | BITMAP AND | |
| 7 | BITMAP MERGE | |
| 8 | BITMAP KEY ITERATION | |
| 9 | TABLE ACCESS FULL | COLORS |
| 10 | BITMAP INDEX RANGE SCAN | CAR_COLOR_IDX |
4-22
Chapter 4
About Adaptive Query Optimization
|- 11 | STATISTICS COLLECTOR | |
|- 12 | BITMAP MERGE | |
|- 13 | BITMAP KEY ITERATION | |
|- 14 | TABLE ACCESS FULL | MODELS |
|- 15 | BITMAP INDEX RANGE SCAN | CAR_MODEL_IDX |
|- 16 | STATISTICS COLLECTOR | |
|- 17 | BITMAP MERGE | |
|- 18 | BITMAP KEY ITERATION | |
|- 19 | TABLE ACCESS FULL | MAKES |
|- 20 | BITMAP INDEX RANGE SCAN | CAR_MAKE_IDX |
| 21 | TABLE ACCESS BY USER ROWID | CARS |
| 22 | MERGE JOIN CARTESIAN | |
| 23 | MERGE JOIN CARTESIAN | |
| 24 | MERGE JOIN CARTESIAN | |
| 25 | TABLE ACCESS FULL | MAKES |
| 26 | BUFFER SORT | |
| 27 | TABLE ACCESS FULL | MODELS |
| 28 | BUFFER SORT | |
| 29 | TABLE ACCESS FULL | COLORS |
| 30 | BUFFER SORT | |
| 31 | TABLE ACCESS FULL | HCC_TAB |
-----------------------------------------------------------
Note
-----
- dynamic statistics used: dynamic sampling (level=2)
- star transformation used for this statement
- this is an adaptive plan (rows marked '-' are inactive)
See Also:
4-23
Chapter 4
About Adaptive Query Optimization
Related Topics
• Supplemental Dynamic Statistics
By default, when optimizer statistics are missing, stale, or insufficient, the
database automatically gathers dynamic statistics during a parse. The database
uses recursive SQL to scan a small random sample of table blocks.
4-24
Chapter 4
About Adaptive Query Optimization
The optimizer uses the information gathered during the previous execution to help
determine an alternative plan. The optimizer can reoptimize a query several times,
each time gathering additional data and further improving the plan.
Automatic reoptimization takes the following forms:
• Reoptimization: Statistics Feedback
A form of reoptimization known as statistics feedback (formerly known as
cardinality feedback) automatically improves plans for repeated queries that have
cardinality misestimates.
• Reoptimization: Performance Feedback
Another form of reoptimization is performance feedback. This reoptimization helps
improve the degree of parallelism automatically chosen for repeated SQL
statements when PARALLEL_DEGREE_POLICY is set to ADAPTIVE.
Related Topics
• Controlling Adaptive Optimization
In Oracle Database, adaptive query optimization is the process by which the
optimizer adapts an execution plan based on statistics collected at run time.
4-25
Chapter 4
About Adaptive Query Optimization
feedback for join cardinality misestimates, but it does not disable statistics feedback
for single-table cardinality misestimates.
Example 4-3 Statistics Feedback
This example shows how the database uses statistics feedback to adjust incorrect
estimates.
1. The user oe runs the following query of the orders, order_items, and
product_information tables:
2. Querying the plan in the cursor shows that the estimated rows (E-Rows) is far
fewer than the actual rows (A-Rows).
---------------------------------------------------------------------------------------
-----------
| Id | Operation | Name |Starts|E-Rows|A-Rows|A-Time|Buffers|OMem|
1Mem|O/1/M|
---------------------------------------------------------------------------------------
-----------
| 0| SELECT STATEMENT | | 1| | 269 |00:00:00.14|1338|
| | |
| 1| NESTED LOOPS | | 1| 1 | 269 |00:00:00.14|1338|
| | |
| 2| MERGE JOIN CARTESIAN| | 1| 4 |9135 |00:00:00.05| 33|
| | |
|*3| TABLE ACCESS FULL |PRODUCT_INFORMATION| 1| 1 | 87 |00:00:00.01| 32|
| | |
| 4| BUFFER SORT | | 87| 105 |9135 |00:00:00.02| 1|4096|
4096|1/0/0|
| 5| INDEX FULL SCAN |ORDER_PK | 1| 105 | 105 |00:00:00.01| 1|
| | |
|*6| INDEX UNIQUE SCAN |ORDER_ITEMS_UK |9135| 1 | 269 |00:00:00.04|1305|
| | |
---------------------------------------------------------------------------------------
-----------
4-26
Chapter 4
About Adaptive Query Optimization
4. Querying the plan in the cursor shows that the optimizer used statistics feedback
(shown in the Note) for the second execution, and also chose a different plan.
---------------------------------------------------------------------------------------
-----------
|Id | Operation | Name | Starts |E-Rows|A-Rows|A-Time|Buffers|Reads|OMem|
1Mem|O/1/M|
---------------------------------------------------------------------------------------
-----------
| 0| SELECT STATEMENT | | 1| | 269 |00:00:00.05|60|1|
| | |
| 1| NESTED LOOPS | | 1|269| 269 |00:00:00.05|60|1|
| | |
|*2| HASH JOIN | | 1|313| 269 |00:00:00.05|39|1|1398K|
1398K|1/0/0|
|*3| TABLE ACCESS FULL |PRODUCT_INFORMATION| 1| 87| 87 |00:00:00.01|15|0|
| | |
| 4| INDEX FAST FULL SCAN|ORDER_ITEMS_UK | 1|665| 665 |00:00:00.01|24|1|
| | |
|*5| INDEX UNIQUE SCAN |ORDER_PK |269| 1| 269 |00:00:00.01|21|0|
| | |
---------------------------------------------------------------------------------------
-----------
2 - access("P"."PRODUCT_ID"="O"."PRODUCT_ID")
3 - filter(("MIN_PRICE"<40 AND "LIST_PRICE"<50))
5 - access("O"."ORDER_ID"="ORDER_ID")
Note
-----
- statistics feedback used for this statement
In the preceding output, the estimated number of rows (269) in Step 1 matches the
actual number of rows.
4-27
Chapter 4
About Adaptive Query Optimization
Note:
Even if PARALLEL_DEGREE_POLICY is not set to ADAPTIVE, statistics feedback
may influence the degree of parallelism chosen for a statement.
See Also:
4-28
Chapter 4
About Approximate Query Processing
Note:
Setting OPTIMIZER_ADAPTIVE_STATISTICS to FALSE preserves statistics
feedback for single-table cardinality misestimates.
See Also:
4-29
Chapter 4
About Approximate Query Processing
For example, a company requires quick detection and response to credit card
fraud.
• Explorative queries of large data sets must be fast.
For example, a user might want to find out a list of departments whose sales have
approximately reached a specific threshold. A user would form targeted queries on
these departments to find more detailed information, such as the exact sales
number, the locations of these departments, and so on.
For large data sets, exact aggregation queries consume extensive memory, often
spilling to temp space, and can be unacceptably slow. Applications are often more
interested in a general pattern than exact results, so customers are willing to sacrifice
exactitude for speed. For example, if the goal is to show a bar chart depicting the most
popular products, then whether a product sold 1 million units or .999 million units is
statistically insignificant.
Oracle Database implements its solution through approximate query processing.
Typically, the accuracy of the approximate aggregation is over 97% (with 95%
confidence), but the processing time is orders of magnitude faster. The database uses
less CPU, and avoids writing to temp files.
You can implement approximate query processing without changing existing code by
using the APPROX_FOR_* initialization parameters. You can set these parameters at the
database or session level. The following table describes initialization parameters and
SQL functions relevant to approximation techniques.
4-30
Chapter 4
About SQL Plan Management
See Also:
"NDV Algorithms: Adaptive Sampling and HyperLogLog"
4-31
Chapter 4
About the Expression Statistics Store (ESS)
See Also:
See Also:
4-32
5
Query Transformations
The optimizer employs many query transformation techniques. This chapter describes
some of the most important.
This chapter contains the following topics:
• OR Expansion
In OR expansion, the optimizer transforms a query block containing top-level
disjunctions into the form of a UNION ALL query that contains two or more
branches. The optimizer achieves this goal by splitting the disjunction into its
components, and then associating each component with a branch of a UNION ALL
query.
• View Merging
In view merging, the optimizer merges the query block representing a view into
the query block that contains it.
• Predicate Pushing
In predicate pushing, the optimizer "pushes" the relevant predicates from the
containing query block into the view query block.
• Subquery Unnesting
In subquery unnesting, the optimizer transforms a nested query into an
equivalent join statement, and then optimizes the join.
• Query Rewrite with Materialized Views
A materialized view is a query result stored in a table.
• Star Transformation
Star transformation is an optimizer transformation that avoids full table scans of
fact tables in a star schema.
• In-Memory Aggregation (VECTOR GROUP BY)
The key optimization of in-memory aggregation is to aggregate while scanning.
• Cursor-Duration Temporary Tables
To materialize the intermediate results of a query, Oracle Database may implicitly
create a cursor-duration temporary table in memory during query compilation.
• Table Expansion
In table expansion, the optimizer generates a plan that uses indexes on the read-
mostly portion of a partitioned table, but not on the active portion of the table.
• Join Factorization
In the cost-based transformation known as join factorization, the optimizer can
factorize common computations from branches of a UNION ALL query.
Related Topics
• Query Transformer
For some statements, the query transformer determines whether it is
advantageous to rewrite the original SQL statement into a semantically equivalent
SQL statement with a lower cost.
5-1
Chapter 5
OR Expansion
5.1 OR Expansion
In OR expansion, the optimizer transforms a query block containing top-level
disjunctions into the form of a UNION ALL query that contains two or more branches.
The optimizer achieves this goal by splitting the disjunction into its components, and
then associating each component with a branch of a UNION ALL query.
The optimizer can choose OR expansion for various reasons. For example, it may
enable more efficient access paths or alternative join methods that avoid Cartesian
products. As always, the optimizer performs the expansion only if the cost of the
transformed statement is lower than the cost of the original statement.
In previous releases, the optimizer used the CONCATENATION operator to perform the OR
expansion. Starting in Oracle Database 12c Release 2 (12.2), the optimizer uses the
UNION-ALL operator instead. The framework provides the following enhancements:
You then connect as the user hr, and execute the following query, which joins the
employees and departments tables:
SELECT *
FROM employees e, departments d
5-2
Chapter 5
OR Expansion
With OR expansion, the optimizer breaks the disjunctive predicate into two independent
predicates, as shown in the following example:
SELECT *
FROM employees e, departments d
WHERE e.email = 'SSTILES'
AND e.department_id = d.department_id
UNION ALL
SELECT *
FROM employees e, departments d
WHERE d.department_name = 'Treasury'
AND e.department_id = d.department_id;
-------------------------------------------------------------------------------------------
----
| Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
-------------------------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT | | | |122
(100)| |
| 1 | VIEW |VW_ORE_19FF4E3E |9102|1679K|122 (5) |
00:00:01|
| 2 | UNION-ALL | | | |
| |
| 3 | NESTED LOOPS | | 1 | 78 | 4 (0) |
00:00:01|
| 4 | TABLE ACCESS BY INDEX ROWID | EMPLOYEES | 1 | 57 | 3 (0) |
00:00:01|
|*5 | INDEX UNIQUE SCAN | EMP_EMAIL_UK | 1 | | 2 (0) |
00:00:01|
| 6 | TABLE ACCESS BY INDEX ROWID | DEPARTMENTS | 1 | 21 | 1 (0) |
00:00:01|
|*7 | INDEX UNIQUE SCAN | DEPT_ID_PK | 1 | | 0 (0)
| |
| 8 | NESTED LOOPS | |9101| 693K|118 (5) |
00:00:01|
| 9 | TABLE ACCESS BY INDEX ROWID | DEPARTMENTS | 1 | 21 | 1 (0) |
00:00:01|
|*10| INDEX UNIQUE SCAN |DEPARTMENT_NAME_UK| 1 | | 0 (0)
5-3
Chapter 5
View Merging
| |
|*11| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |9101| 506K|117 (5) |
00:00:01|
|*12| INDEX RANGE SCAN |EMP_DEPARTMENT_IX |9101| | 35 (6) |
00:00:01|
-------------------------------------------------------------------------------------------
----
5 - access("E"."EMAIL"='SSTILES')
7 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
10 - access("D"."DEPARTMENT_NAME"='Treasury')
11 - filter(LNNVL("E"."EMAIL"='SSTILES'))
12 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
35 rows selected.
Note:
You can use hints to override view merging rejected because of cost or
heuristics, but not validity.
5-4
Chapter 5
View Merging
See Also:
Simple view merging frequently results in a more optimal plan because of the
additional join orders and access paths available after the merge. A view may not be
valid for simple view merging because:
• The view contains constructs not included in select-project-join views, including:
– GROUP BY
– DISTINCT
– Outer join
– MODEL
5-5
Chapter 5
View Merging
– CONNECT BY
– Set operators
– Aggregation
• The view appears on the right side of a semijoin or antijoin.
• The view contains subqueries in the SELECT list.
• The outer query block contains PL/SQL functions.
• The view participates in an outer join, and does not meet one of the several
additional validity requirements that determine whether the view can be merged.
Example 5-2 Simple View Merging
The following query joins the hr.employees table with the dept_locs_v view, which
returns the street address for each department. dept_locs_v is a join of the
departments and locations tables.
The database can execute the preceding query by joining departments and locations
to generate the rows of the view, and then joining this result to employees. Because
the query contains the view dept_locs_v, and this view contains two tables, the
optimizer must use one of the following join orders:
• employees, dept_locs_v (departments, locations)
• employees, dept_locs_v (locations, departments)
• dept_locs_v (departments, locations), employees
• dept_locs_v (locations, departments), employees
Join methods are also constrained. The index-based nested loops join is not feasible
for join orders that begin with employees because no index exists on the column from
this view. Without view merging, the optimizer generates the following execution plan:
-----------------------------------------------------------------
| Id | Operation | Name | Cost (%CPU)|
-----------------------------------------------------------------
| 0 | SELECT STATEMENT | | 7 (15)|
|* 1 | HASH JOIN | | 7 (15)|
| 2 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES | 2 (0)|
|* 3 | INDEX RANGE SCAN | EMP_NAME_IX | 1 (0)|
| 4 | VIEW | | 5 (20)|
|* 5 | HASH JOIN | | 5 (20)|
| 6 | TABLE ACCESS FULL | LOCATIONS | 2 (0)|
| 7 | TABLE ACCESS FULL | DEPARTMENTS | 2 (0)|
-----------------------------------------------------------------
5-6
Chapter 5
View Merging
View merging merges the tables from the view into the outer query block, removing the
inner query block. After view merging, the query is as follows:
Because all three tables appear in one query block, the optimizer can choose from the
following six join orders:
• employees, departments, locations
• employees, locations, departments
• departments, employees, locations
• departments, locations, employees
• locations, employees, departments
• locations, departments, employees
The joins to employees and departments can now be index-based. After view merging,
the optimizer chooses the following more efficient plan, which uses nested loops:
-------------------------------------------------------------------
| Id | Operation | Name | Cost (%CPU)|
-------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 4 (0)|
| 1 | NESTED LOOPS | | |
| 2 | NESTED LOOPS | | 4 (0)|
| 3 | NESTED LOOPS | | 3 (0)|
| 4 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES | 2 (0)|
|* 5 | INDEX RANGE SCAN | EMP_NAME_IX | 1 (0)|
| 6 | TABLE ACCESS BY INDEX ROWID| DEPARTMENTS | 1 (0)|
|* 7 | INDEX UNIQUE SCAN | DEPT_ID_PK | 0 (0)|
|* 8 | INDEX UNIQUE SCAN | LOC_ID_PK | 0 (0)|
| 9 | TABLE ACCESS BY INDEX ROWID | LOCATIONS | 1 (0)|
-------------------------------------------------------------------
5-7
Chapter 5
View Merging
See Also:
The Oracle Optimizer blog at https://blogs.oracle.com/optimizer/ to
learn about outer join view merging, which is a special case of simple view
merging
The following query finds all of the customers from the United States who have bought
at least 100 fur-trimmed sweaters:
5-8
Chapter 5
View Merging
The cust_prod_totals_v view is eligible for complex view merging. After merging, the
query is as follows:
The transformed query is cheaper than the untransformed query, so the optimizer
chooses to merge the view. In the untransformed query, the GROUP BY operator applies
to the entire sales table in the view. In the transformed query, the joins to products
and customers filter out a large portion of the rows from the sales table, so the GROUP
BY operation is lower cost. The join is more expensive because the sales table has not
been reduced, but it is not much more expensive because the GROUP BY operation
does not reduce the size of the row set very much in the original query. If any of the
preceding characteristics were to change, merging the view might no longer be lower
cost. The final plan, which does not include a view, is as follows:
--------------------------------------------------------
| Id | Operation | Name | Cost (%CPU)|
--------------------------------------------------------
| 0 | SELECT STATEMENT | | 2101 (18)|
|* 1 | FILTER | | |
| 2 | HASH GROUP BY | | 2101 (18)|
|* 3 | HASH JOIN | | 2099 (18)|
|* 4 | HASH JOIN | | 1801 (19)|
|* 5 | TABLE ACCESS FULL| PRODUCTS | 96 (5)|
| 6 | TABLE ACCESS FULL| SALES | 1620 (15)|
|* 7 | TABLE ACCESS FULL | CUSTOMERS | 296 (11)|
--------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter(SUM("QUANTITY_SOLD")>100)
3 - access("C"."CUST_ID"="CUST_ID")
4 - access("PROD_ID"="P"."PROD_ID")
5 - filter("P"."PROD_NAME"='T3 Faux Fur-Trimmed Sweater')
7 - filter("C"."COUNTRY_ID"='US')
5-9
Chapter 5
View Merging
After determining that view merging produces a lower-cost plan, the optimizer rewrites
the query into this equivalent query:
-------------------------------------------
| Id | Operation | Name |
-------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | VIEW | VM_NWVW_1 |
| 2 | HASH UNIQUE | |
|* 3 | HASH JOIN | |
|* 4 | HASH JOIN | |
|* 5 | TABLE ACCESS FULL| PRODUCTS |
| 6 | TABLE ACCESS FULL| SALES |
|* 7 | TABLE ACCESS FULL | CUSTOMERS |
-------------------------------------------
The preceding plan contains a view named vm_nwvw_1, known as a projection view,
even after view merging has occurred. Projection views appear in queries in which a
DISTINCT view has been merged, or a GROUP BY view is merged into an outer query
block that also contains GROUP BY, HAVING, or aggregates. In the latter case, the
projection view contains the GROUP BY, HAVING, and aggregates from the original outer
query block.
In the preceding example of a projection view, when the optimizer merges the view, it
moves the DISTINCT operator to the outer query block, and then adds several
additional columns to maintain semantic equivalence with the original query.
Afterward, the query can select only the desired columns in the SELECT list of the outer
query block. The optimization retains all of the benefits of view merging: all tables are
in one query block, the optimizer can permute them as needed in the final join order,
and the DISTINCT operation has been delayed until after all of the joins complete.
5-10
Chapter 5
Predicate Pushing
You create a view that references employees and contract_workers. The view is
defined with a query that uses the UNION set operator, as follows:
SELECT last_name
FROM all_employees_vw
WHERE department_id = 50;
Because the view is a UNION set query, the optimizer cannot merge the view's query
into the accessing query block. Instead, the optimizer can transform the accessing
statement by pushing its predicate, the WHERE clause condition department_id=50, into
the view's UNION set query. The equivalent transformed query is as follows:
SELECT last_name
FROM ( SELECT employee_id, last_name, job_id, commission_pct,
department_id
FROM employees
WHERE department_id=50
UNION
SELECT employee_id, last_name, job_id, commission_pct,
department_id
5-11
Chapter 5
Subquery Unnesting
FROM contract_workers
WHERE department_id=50 );
The transformed query can now consider index access in each of the query blocks.
For example, suppose you connect as user sh and execute the following query:
SELECT *
FROM sales
WHERE cust_id IN ( SELECT cust_id
FROM customers );
Because the customers.cust_id column is a primary key, the optimizer can transform
the complex query into the following join statement that is guaranteed to return the
same data:
SELECT sales.*
FROM sales, customers
WHERE sales.cust_id = customers.cust_id;
If the optimizer cannot transform a complex statement into a join statement, it selects
execution plans for the parent statement and the subquery as though they were
separate statements. The optimizer then executes the subquery and uses the rows
returned to execute the parent query. To improve execution speed of the overall
execution plan, the optimizer orders the subplans efficiently.
5-12
Chapter 5
Star Transformation
See Also:
Oracle Database Data Warehousing Guide to learn more about query rewrite
5-13
Chapter 5
Star Transformation
products times
sales
(amount_sold,
quantity_sold)
Fact Table
customers channels
A snowflake schema is a star schema in which the dimension tables reference other
tables. A snowstorm schema is a combination of snowflake schemas.
See Also:
Oracle Database Data Warehousing Guide to learn more about star
schemas
5-14
Chapter 5
Star Transformation
See Also:
Oracle Database Reference to learn about the
STAR_TRANSFORMATION_ENABLED initialization parameter
SELECT c.cust_city,
t.calendar_quarter_desc,
SUM(s.amount_sold) sales_amount
FROM sales s,
times t,
customers c,
channels ch
WHERE s.time_id = t.time_id
AND s.cust_id = c.cust_id
AND s.channel_id = ch.channel_id
AND c.cust_state_province = 'CA'
AND ch.channel_desc = 'Internet'
AND t.calendar_quarter_desc IN ('1999-01','1999-02')
GROUP BY c.cust_city, t.calendar_quarter_desc;
5-15
Chapter 5
Star Transformation
In this example, sales is the fact table, and the other tables are dimension tables. The
sales table contains one row for every sale of a product, so it could conceivably
contain billions of sales records. However, only a few products are sold to customers
in California through the Internet for the specified quarters.
Example 5-6 Star Transformation
This example shows a star transformation of the query in Example 5-5. The
transformation avoids a full table scan of sales.
Each bit in the bitmap corresponds to a row in the fact table. The bit is set when the
key value from the subquery is same as the value in the row of the fact table. For
example, in the bitmap 101000... (the ellipses indicates that the values for the
remaining rows are 0), rows 1 and 3 of the fact table have matching key values from
the subquery.
5-16
Chapter 5
Star Transformation
The operations in lines 12, 17, and 22 iterate over the keys from the subqueries and
retrieve the corresponding bitmaps. In Example 5-6, the customers subquery seeks
the IDs of customers whose state or province is CA. Assume that the bitmap 101000...
corresponds to the customer ID key value 103515 from the customers table subquery.
Also assume that the customers subquery produces the key value 103516 with the
bitmap 010000..., which means that only row 2 in sales has a matching key value
from the subquery.
The database merges (using the OR operator) the bitmaps for each subquery (lines 11,
16, 21). In our customers example, the database produces a single bitmap 111000...
for the customers subquery after merging the two bitmaps:
In line 10, the database applies the AND operator to the merged bitmaps. Assume that
after the database has performed all OR operations, the resulting bitmap for channels
is 100000... If the database performs an AND operation on this bitmap and the bitmap
from customers subquery, then the result is as follows:
In line 9, the database generates the corresponding rowids of the final bitmap. The
database retrieves rows from the sales fact table using the rowids (line 26). In our
example, the database generate only one rowid, which corresponds to the first row,
and thus fetches only a single row instead of scanning the entire sales table.
---------------------------------------------------------------------------
----
| Id | Operation | Name
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT |
| 1 | HASH GROUP BY |
|* 2 | HASH JOIN |
|* 3 | TABLE ACCESS FULL | CUSTOMERS
|* 4 | HASH JOIN |
|* 5 | TABLE ACCESS FULL | TIMES
| 6 | VIEW | VW_ST_B1772830
| 7 | NESTED LOOPS |
| 8 | PARTITION RANGE SUBQUERY |
| 9 | BITMAP CONVERSION TO ROWIDS|
| 10 | BITMAP AND |
| 11 | BITMAP MERGE |
| 12 | BITMAP KEY ITERATION |
| 13 | BUFFER SORT |
|* 14 | TABLE ACCESS FULL | CHANNELS
|* 15 | BITMAP INDEX RANGE SCAN| SALES_CHANNEL_BIX
5-17
Chapter 5
Star Transformation
| 16 | BITMAP MERGE |
| 17 | BITMAP KEY ITERATION |
| 18 | BUFFER SORT |
|* 19 | TABLE ACCESS FULL | TIMES
|* 20 | BITMAP INDEX RANGE SCAN| SALES_TIME_BIX
| 21 | BITMAP MERGE |
| 22 | BITMAP KEY ITERATION |
| 23 | BUFFER SORT |
|* 24 | TABLE ACCESS FULL | CUSTOMERS
|* 25 | BITMAP INDEX RANGE SCAN| SALES_CUST_BIX
| 26 | TABLE ACCESS BY USER ROWID | SALES
---------------------------------------------------------------------------
----
2 - access("ITEM_1"="C"."CUST_ID")
3 - filter("C"."CUST_STATE_PROVINCE"='CA')
4 - access("ITEM_2"="T"."TIME_ID")
5 - filter(("T"."CALENDAR_QUARTER_DESC"='1999-01'
OR "T"."CALENDAR_QUARTER_DESC"='1999-02'))
14 - filter("CH"."CHANNEL_DESC"='Internet')
15 - access("S"."CHANNEL_ID"="CH"."CHANNEL_ID")
19 - filter(("T"."CALENDAR_QUARTER_DESC"='1999-01'
OR "T"."CALENDAR_QUARTER_DESC"='1999-02'))
20 - access("S"."TIME_ID"="T"."TIME_ID")
24 - filter("C"."CUST_STATE_PROVINCE"='CA')
25 - access("S"."CUST_ID"="C"."CUST_ID")
Note
-----
- star transformation used for this statement
If the optimizer cannot eliminate the join back, however, then the database stores the
subquery results in a temporary table to avoid rescanning the dimension table for
bitmap key generation and join back. Also, if the query runs in parallel, then the
database materializes the results so that each parallel execution server can select the
results from the temporary table instead of executing the subquery again.
Example 5-8 Star Transformation Using Temporary Table
In this example, the database materializes the results of the subquery on customers
into a temporary table:
5-18
Chapter 5
Star Transformation
Example 5-9 Partial Execution Plan for Star Transformation Using Temporary
Table
The following example shows an edited version of the execution plan for the query in
Example 5-8:
---------------------------------------------------------------------------
----
| Id | Operation | Name
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT |
| 1 | TEMP TABLE TRANSFORMATION |
| 2 | LOAD AS SELECT |
|* 3 | TABLE ACCESS FULL | CUSTOMERS
| 4 | HASH GROUP BY |
|* 5 | HASH JOIN |
| 6 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6613_C716F
|* 7 | HASH JOIN |
|* 8 | TABLE ACCESS FULL | TIMES
| 9 | VIEW | VW_ST_A3F94988
| 10 | NESTED LOOPS |
| 11 | PARTITION RANGE SUBQUERY |
| 12 | BITMAP CONVERSION TO ROWIDS|
| 13 | BITMAP AND |
| 14 | BITMAP MERGE |
| 15 | BITMAP KEY ITERATION |
| 16 | BUFFER SORT |
|* 17 | TABLE ACCESS FULL | CHANNELS
|* 18 | BITMAP INDEX RANGE SCAN| SALES_CHANNEL_BIX
5-19
Chapter 5
In-Memory Aggregation (VECTOR GROUP BY)
| 19 | BITMAP MERGE |
| 20 | BITMAP KEY ITERATION |
| 21 | BUFFER SORT |
|* 22 | TABLE ACCESS FULL | TIMES
|* 23 | BITMAP INDEX RANGE SCAN| SALES_TIME_BIX
| 24 | BITMAP MERGE |
| 25 | BITMAP KEY ITERATION |
| 26 | BUFFER SORT |
| 27 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6613_C716F
|* 28 | BITMAP INDEX RANGE SCAN| SALES_CUST_BIX
| 29 | TABLE ACCESS BY USER ROWID | SALES
---------------------------------------------------------------------------
----
3 - filter("C"."CUST_STATE_PROVINCE"='CA')
5 - access("ITEM_1"="C0")
7 - access("ITEM_2"="T"."TIME_ID")
8 - filter(("T"."CALENDAR_QUARTER_DESC"='1999-01' OR
"T"."CALENDAR_QUARTER_DESC"='1999-02'))
17 - filter("CH"."CHANNEL_DESC"='Internet')
18 - access("S"."CHANNEL_ID"="CH"."CHANNEL_ID")
22 - filter(("T"."CALENDAR_QUARTER_DESC"='1999-01' OR
"T"."CALENDAR_QUARTER_DESC"='1999-02'))
23 - access("S"."TIME_ID"="T"."TIME_ID")
28 - access("S"."CUST_ID"="C0")
Lines 1, 2, and 3 of the plan materialize the customers subquery into the temporary
table. In line 6, the database scans the temporary table (instead of the subquery) to
build the bitmap from the fact table. Line 27 scans the temporary table for joining back
instead of scanning customers. The database does not need to apply the filter on
customers on the temporary table because the filter is applied while materializing the
temporary table.
See Also:
Oracle Database In-Memory Guide to learn more about in-memory
aggregation
5-20
Chapter 5
Cursor-Duration Temporary Tables
5-21
Chapter 5
Cursor-Duration Temporary Tables
Note:
The metadata for the cursor-duration temporary table stays in memory as
long as the cursor is in memory. The metadata is not stored in the data
dictionary, which means it is not visible through data dictionary views. You
cannot drop the metadata explicitly.
The preceding scenario depends on the availability of memory. For serial queries, the
temporary tables use PGA memory.
The implementation of cursor-duration temporary tables is similar to sorts. If no more
memory is available, then the database writes data to temporary segments. For
cursor-duration temporary tables, the differences are as follows:
• The database releases memory and temporary segments at the end of the query
rather than when the row source is no longer active.
• Data in memory stays in memory, unlike in sorts where data can move between
memory and temporary segments.
When the database uses cursor-duration temporary tables, the keyword CURSOR
DURATION MEMORY appears in the execution plan.
WITH
q1 AS (SELECT department_id, SUM(salary) sum_sal FROM hr.employees GROUP
BY department_id),
q2 AS (SELECT * FROM q1),
q3 AS (SELECT department_id, sum_sal FROM q1)
SELECT * FROM q1
UNION ALL
SELECT * FROM q2
UNION ALL
SELECT * FROM q3;
PLAN_TABLE_OUTPUT
-------------------------------------------------------------------------------------------
-
| Id | Operation | Name |Rows |Cost
(%CPU)|
-------------------------------------------------------------------------------------------
-
| 0 | SELECT STATEMENT | | |6
5-22
Chapter 5
Table Expansion
(100)|
| 1 | TEMP TABLE TRANSFORMATION | | |
|
| 2 | LOAD AS SELECT (CURSOR DURATION MEMORY) | SYS_TEMP_0FD9D6606_1AE004 | |
|
| 3 | HASH GROUP BY | | 11 | 3
(34)|
| 4 | TABLE ACCESS FULL | EMPLOYEES | 107 | 2 (0)
|
| 5 | UNION-ALL | | |
|
| 6 | VIEW | | 11 | 2 (0)
|
| 7 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6606_1AE004 | 11 | 2 (0)
|
| 8 | VIEW | | 11 | 2 (0)
|
| 9 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6606_1AE004 | 11 | 2 (0)
|
| 10 | VIEW | | 11 | 2 (0)
|
| 11 | TABLE ACCESS FULL | SYS_TEMP_0FD9D6606_1AE004 | 11 | 2 (0)
|
-------------------------------------------------------------------------------------------
-
In the preceding plan, TEMP TABLE TRANSFORMATION in Step 1 indicates that the
database used cursor-duration temporary tables to execute the query. The CURSOR
DURATION MEMORY keyword in Step 2 indicates that the database used memory, if
available, to store the results of SYS_TEMP_0FD9D6606_1AE004. If memory was
unavailable, then the database wrote the temporary data to disk.
5-23
Chapter 5
Table Expansion
See Also:
5-24
Chapter 5
Table Expansion
Assumptions
This scenario assumes the following:
• You want to run a star query against the sh.sales table, which is range-partitioned
on the time_id column.
• You want to disable indexes on specific partitions to see the benefits of table
expansion.
SELECT *
FROM sales
WHERE time_id >= TO_DATE('2000-01-01 00:00:00', 'SYYYY-MM-DD
HH24:MI:SS')
AND prod_id = 38;
As shown in the Pstart and Pstop columns in the following plan, the optimizer
determines from the filter that only 16 of the 28 partitions in the table must be
accessed:
------------------------------------------------------------------------
--
|Id| Operation | Name |Pstart|
Pstop|
------------------------------------------------------------------------
--
| 0| SELECT STATEMENT | |
| |
| 1| PARTITION RANGE ITERATOR | | 13 |
28 |
| 2| TABLE ACCESS BY LOCAL INDEX ROWID BATCHED| SALES | 13 |
28 |
| 3| BITMAP CONVERSION TO ROWIDS | |
5-25
Chapter 5
Table Expansion
| |
|*4| BITMAP INDEX SINGLE VALUE |SALES_PROD_BIX| 13 |
28 |
------------------------------------------------------------------------
--
4 - access("PROD_ID"=38)
After the optimizer has determined the partitions to be accessed, it considers any
index that is usable on all of those partitions. In the preceding plan, the optimizer
chose to use the sales_prod_bix bitmap index.
4. Disable the index on the SALES_1995 partition of the sales table:
The preceding DDL disables the index on partition 1, which contains all sales from
before 1996.
Note:
You can obtain the partition information by querying the
USER_IND_PARTITIONS view.
5. Execute the query of sales again, and then query DBMS_XPLAN to obtain the plan.
The output shows that the plan did not change:
------------------------------------------------------------------------
---
|Id| Operation | Name |Pstart|
Pstop
------------------------------------------------------------------------
---
| 0| SELECT STATEMENT | |
| |
| 1| PARTITION RANGE ITERATOR | | 13 |
28 |
| 2| TABLE ACCESS BY LOCAL INDEX ROWID BATCHED| SALES | 13 |
28 |
| 3| BITMAP CONVERSION TO ROWIDS | |
| |
|*4| BITMAP INDEX SINGLE VALUE | SALES_PROD_BIX| 13 |
28 |
------------------------------------------------------------------------
---
5-26
Chapter 5
Table Expansion
---------------------------------------------------
4 - access("PROD_ID"=38)
The plan is the same because the disabled index partition is not relevant to the
query. If all partitions that the query accesses are indexed, then the database can
answer the query using the index. Because the query only accesses partitions 16
through 28, disabling the index on partition 1 does not affect the plan.
6. Disable the indexes for partition 28 (SALES_Q4_2003), which is a partition that the
query needs to access:
By disabling the indexes on a partition that the query does need to access, the
query can no longer use this index (without table expansion).
7. Query the plan using DBMS_XPLAN.
As shown in the following plan, the optimizer does not use the index:
------------------------------------------------------------------------
---
| Id| Operation | Name |Pstart|
Pstop
------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | |
| |
| 1 | PARTITION RANGE ITERATOR | | 13 |
28 |
|*2 | TABLE ACCESS FULL | SALES | 13 |
28 |
------------------------------------------------------------------------
---
SELECT *
FROM sales
WHERE time_id >= TO_DATE('2000-01-01 00:00:00', 'SYYYY-MM-DD
HH24:MI:SS')
AND time_id < TO_DATE('2003-10-01 00:00:00', 'SYYYY-MM-DD
5-27
Chapter 5
Table Expansion
HH24:MI:SS')
AND prod_id = 38
UNION ALL
SELECT *
FROM sales
WHERE time_id >= TO_DATE('2003-10-01 00:00:00', 'SYYYY-MM-DD
HH24:MI:SS')
AND time_id < TO_DATE('2004-01-01 00:00:00', 'SYYYY-MM-DD
HH24:MI:SS')
AND prod_id = 38;
In the preceding query, the first query block in the UNION ALL accesses the
partitions that are indexed, while the second query block accesses the partition
that is not. The two subqueries enable the optimizer to choose to use the index in
the first query block, if it is more optimal than using a table scan of all of the
partitions that are accessed.
9. Query the plan using DBMS_XPLAN.
The plan appears as follows:
------------------------------------------------------------------------
---
|Id| Operation | Name |Pstart|
Pstop|
------------------------------------------------------------------------
---
| 0| SELECT STATEMENT | |
| |
| 1| VIEW | VW_TE_2 |
| |
| 2| UNION-ALL | |
| |
| 3| PARTITION RANGE ITERATOR | | 13|
27|
| 4| TABLE ACCESS BY LOCAL INDEX ROWID BATCHED| SALES | 13|
27|
| 5| BITMAP CONVERSION TO ROWIDS | |
| |
|*6| BITMAP INDEX SINGLE VALUE | SALES_PROD_BIX| 13|
27|
| 7| PARTITION RANGE SINGLE | | 28|
28|
|*8| TABLE ACCESS FULL | SALES | 28|
28|
------------------------------------------------------------------------
---
6 - access("PROD_ID"=38)
8 - filter("PROD_ID"=38)
5-28
Chapter 5
Table Expansion
As shown in the preceding plan, the optimizer uses a UNION ALL for two query
blocks (Step 2). The optimizer chooses an index to access partitions 13 to 27 in
the first query block (Step 6). Because no index is available for partition 28, the
optimizer chooses a full table scan in the second query block (Step 8).
Assumptions
This scenario assumes the following:
• You query the same schema used in "Star Transformation: Scenario".
• The last partition of sales is actively being updated, as is often the case with time-
partitioned tables.
• You want the optimizer to take advantage of table expansion.
3. Query the cursor using DBMS_XPLAN, which shows the following plan:
------------------------------------------------------------------------
---
|Id| Operation | Name | Pstart|
Pstop |
------------------------------------------------------------------------
---
| 0| SELECT STATEMENT | |
| |
| 1| HASH GROUP BY | |
5-29
Chapter 5
Table Expansion
| |
| 2| VIEW |VW_TE_14 |
| |
| 3| UNION-ALL | |
| |
| 4| HASH JOIN | |
| |
| 5| TABLE ACCESS FULL |TIMES |
| |
| 6| VIEW |VW_ST_1319B6D8 |
| |
| 7| NESTED LOOPS | |
| |
| 8| PARTITION RANGE SUBQUERY | |KEY(SQ)|
KEY(SQ)|
| 9| BITMAP CONVERSION TO ROWIDS| |
| |
|10| BITMAP AND | |
| |
|11| BITMAP MERGE | |
| |
|12| BITMAP KEY ITERATION | |
| |
|13| BUFFER SORT | |
| |
|14| TABLE ACCESS FULL |CHANNELS |
| |
|15| BITMAP INDEX RANGE SCAN|SALES_CHANNEL_BIX|KEY(SQ)|
KEY(SQ)|
|16| BITMAP MERGE | |
| |
|17| BITMAP KEY ITERATION | |
| |
|18| BUFFER SORT | |
| |
|19| TABLE ACCESS FULL |TIMES |
| |
|20| BITMAP INDEX RANGE SCAN|SALES_TIME_BIX |KEY(SQ)|
KEY(SQ)|
|21| BITMAP MERGE | |
| |
|22| BITMAP KEY ITERATION | |
| |
|23| BUFFER SORT | |
| |
|24| TABLE ACCESS FULL |CUSTOMERS |
| |
|25| BITMAP INDEX RANGE SCAN|SALES_CUST_BIX |KEY(SQ)|
KEY(SQ)|
|26| TABLE ACCESS BY USER ROWID |SALES | ROWID |
ROWID |
|27| NESTED LOOPS | |
| |
|28| NESTED LOOPS | |
| |
5-30
Chapter 5
Join Factorization
The preceding plan uses table expansion. The UNION ALL branch that is accessing
every partition except the last partition uses star transformation. Because the
indexes on partition 28 are disabled, the database accesses the final partition
using a full table scan.
Related Topics
• Star Transformation
Star transformation is an optimizer transformation that avoids full table scans of
fact tables in a star schema.
5-31
Chapter 5
Join Factorization
In the preceding query, table t1 appears in both UNION ALL branches, as does the filter
predicate t1.c1 > 1 and the join predicate t1.c1 = t2.c1. Without any
transformation, the database must perform the scan and the filtering on table t1 twice,
one time for each branch.
Example 5-11 Factorized Query
Example 5-10
5-32
Chapter 5
Join Factorization
In this case, because table t1 is factorized, the database performs the table scan and
the filtering on t1 only one time. If t1 is large, then this factorization avoids the huge
performance cost of scanning and filtering t1 twice.
Note:
If the branches in a UNION ALL query have clauses that use the DISTINCT
function, then join factorization is not valid.
SELECT *
FROM t5, (SELECT t1.c1, t2.c2
FROM t1, t2, t3
WHERE t1.c1 = t2.c1
AND t1.c1 > 1
AND t2.c2 = 2
AND t2.c2 = t3.c2
UNION ALL
SELECT t1.c1, t2.c2
FROM t1, t2, t4
WHERE t1.c1 = t2.c1
AND t1.c1 > 1
AND t2.c3 = t4.c3) V
WHERE t5.c1 = V.c1
t1t2t3t5
Example 5-13 Factorization of t1 from View V
If join factorization factorizes t1 from view V, as shown in the following query, then the
database can join t1 with t5.:
SELECT *
FROM t5, ( SELECT t1.c1, VW_JF_1.item_2
FROM t1, (SELECT t2.c1 item_1, t2.c2 item_2
FROM t2, t3
WHERE t2.c2 = t3.c2
AND t2.c2 = 2
UNION ALL
SELECT t2.c1 item_1, t2.c2 item_2
FROM t2, t4
5-33
Chapter 5
Join Factorization
The preceding query transformation opens up new join orders. However, join
factorization imposes specific join orders. For example, in the preceding query, tables
t2 and t3 appear in the first branch of the UNION ALL query in view VW_JF_1. The
database must join t2 with t3 before it can join with t1, which is not defined within the
VW_JF_1 view. The imposed join order may not necessarily be the best join order. For
this reason, the optimizer performs join factorization using the cost-based
transformation framework. The optimizer calculates the cost of the plans with and
without join factorization, and then chooses the cheapest plan.
Example 5-14 Factorization of t1 from View V with View Definition Removed
The following query is the same query in Example 5-13, but with the view definition
removed so that the factorization is easier to see:
SELECT *
FROM t5, (SELECT t1.c1, VW_JF_1.item_2
FROM t1, VW_JF_1
WHERE t1.c1 = VW_JF_1.item_1
AND t1.c1 > 1)
WHERE t5.c1 = V.c1
The following example shows the transformation. Table t2 now no longer appears in
the UNION ALL branches of the subquery.
5-34
Chapter 5
Join Factorization
FROM t1
WHERE t1.c1 = 2) VW_JF_1
WHERE VW_JF_1.item_1 = t2.c1(+)
5-35
Part III
Query Execution Plans
If a query has suboptimal performance, the execution plan is the key tool for
understanding the problem and supplying a solution.
This part contains the following chapters:
• Generating and Displaying Execution Plans
A thorough understanding of execution plans is essential to SQL tuning.
• Reading Execution Plans
Execution plans are represented as a tree of operations.
6
Generating and Displaying Execution Plans
A thorough understanding of execution plans is essential to SQL tuning.
This chapter contains the following topics:
• Introduction to Execution Plans
The combination of the steps that Oracle Database uses to execute a statement is
an execution plan.
• About Plan Generation and Display
The EXPLAIN PLAN statement displays execution plans that the optimizer chooses
for SELECT, UPDATE, INSERT, and DELETE statements.
• Generating Plan Output Using the EXPLAIN PLAN Statement
The EXPLAIN PLAN statement enables you to examine the execution plan that the
optimizer chose for a SQL statement.
• Displaying PLAN_TABLE Output
You can use scripts or a package to display the plan output.
6-1
Chapter 6
About Plan Generation and Display
See Also:
6-2
Chapter 6
About Plan Generation and Display
EXPLAIN PLAN output shows how the database would run the SQL statement when the
statement was explained. This plan can differ from the actual execution plan a SQL
statement uses because of differences in the execution environment and explain plan
environment.
Note:
To avoid possible SQL performance regression that may result from
execution plan changes, consider using SQL plan management.
• Different Schemas
Schemas can differ for various reasons.
• Different Costs
Even if the schemas are the same, the optimizer can choose different execution
plans when the costs are different.
See Also:
6-3
Chapter 6
About Plan Generation and Display
6-4
Chapter 6
About Plan Generation and Display
V$SQL_PLAN contains the execution plan for every statement stored in the shared SQL
area. Its definition is similar to PLAN_TABLE.
The advantage of V$SQL_PLAN over EXPLAIN PLAN is that you do not need to know the
compilation environment that was used to execute a particular statement. For EXPLAIN
PLAN, you would need to set up an identical environment to get the same plan when
executing the statement.
The V$SQL_PLAN_STATISTICS view provides the actual execution statistics for every
operation in the plan, such as the number of output rows and elapsed time. All
statistics, except the number of output rows, are cumulative. For example, the
statistics for a join operation also includes the statistics for its two inputs. The statistics
in V$SQL_PLAN_STATISTICS are available for cursors that have been compiled with the
STATISTICS_LEVEL initialization parameter set to ALL.
See Also:
• "PLAN_TABLE Columns"
• "Monitoring Database Operations " for information about the
V$SQL_PLAN_MONITOR view
• Oracle Database Reference for more information about V$SQL_PLAN
views
• Oracle Database Reference for information about the STATISTICS_LEVEL
initialization parameter
6-5
Chapter 6
About Plan Generation and Display
See Also:
For example, start a SQL*Plus session, connect with SYSDBA privileges, and run the
script as follows:
@$ORACLE_HOME/rdbms/admin/catplan.sql
Oracle recommends that you drop and rebuild your local PLAN_TABLE table after
upgrading the version of the database because the columns might change. This can
cause scripts to fail or cause TKPROF to fail, if you are specifying the table.
If you do not want to use the name PLAN_TABLE, create a new synonym after running
the catplan.sql script. For example:
See Also:
6-6
Chapter 6
Generating Plan Output Using the EXPLAIN PLAN Statement
Prerequisites
This task assumes that a sample output table named PLAN_TABLE exists in your
schema. If this table does not exist, then run the SQL script catplan.sql.
• You must have the privileges necessary to insert rows into an existing output table
that you specify to hold the execution plan
• You must also have the privileges necessary to execute the SQL statement for
which you are determining the execution plan. If the SQL statement accesses a
view, then you must have privileges to access any tables and views on which the
view is based. If the view is based on another view that is based on a table, then
you must have privileges to access both the other view and its underlying table.
To examine the execution plan produced by an EXPLAIN PLAN statement, you must
have the privileges necessary to query the output table.
To explain a statement:
1. Start SQL*Plus or SQL Developer, and log in to the database as a user with the
requisite permissions.
2. Include the EXPLAIN PLAN FOR clause immediately before the SQL statement.
The following example explains the plan for a query of the employees table:
6-7
Chapter 6
Generating Plan Output Using the EXPLAIN PLAN Statement
3. After issuing the EXPLAIN PLAN statement, use a script or package provided by
Oracle Database to display the most recent plan table output.
The following example uses the DBMS_XPLAN.DISPLAY function:
------------------------------------------------------------------------
-----------
| Id | Operation | Name | Rows | Bytes | Cost
(%CPU)| Time |
------------------------------------------------------------------------
-----------
| 0 | SELECT STATEMENT | | 4 | 124 | 5
(20)| 00:00:01 |
| 1 | SORT ORDER BY | | 4 | 124 | 5
(20)| 00:00:01 |
|* 2 | HASH JOIN | | 4 | 124 | 4
(0)| 00:00:01 |
|* 3 | TABLE ACCESS FULL| EMPLOYEES | 4 | 60 | 2
(0)| 00:00:01 |
| 4 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2
(0)| 00:00:01 |
------------------------------------------------------------------------
-----------
1 - SEL$1
3 - SEL$1 / E@SEL$1
4 - SEL$1 / D@SEL$1
2 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
3 - filter("SALARY"<3000)
1 - (#keys=1) INTERNAL_FUNCTION("E"."SALARY")[22],
"E"."LAST_NAME"[VARCHAR2,25], "D"."DEPARTMENT_NAME"[VARCHAR2,30]
2 - (#keys=1) "E"."LAST_NAME"[VARCHAR2,25], "SALARY"[NUMBER,22],
"D"."DEPARTMENT_NAME"[VARCHAR2,30],
"D"."DEPARTMENT_NAME"[VARCHAR2,30]
3 - "E"."LAST_NAME"[VARCHAR2,25], "SALARY"[NUMBER,22],
"E"."DEPARTMENT_ID"[NUMBER,22]
6-8
Chapter 6
Generating Plan Output Using the EXPLAIN PLAN Statement
4 - "D"."DEPARTMENT_ID"[NUMBER,22],
"D"."DEPARTMENT_NAME"[VARCHAR2,30]
Note
-----
- this is an adaptive plan
The execution order in EXPLAIN PLAN output begins with the line that is the furthest
indented to the right. The next step is the parent of that line. If two lines are
indented equally, then the top line is normally executed first.
Note:
The steps in the EXPLAIN PLAN output in this chapter may be different on
your database. The optimizer may choose different execution plans,
depending on database configurations.
See Also:
EXPLAIN PLAN
SET STATEMENT_ID = 'st1' FOR
SELECT last_name FROM employees;
6-9
Chapter 6
Displaying PLAN_TABLE Output
EXPLAIN PLAN
INTO my_plan_table FOR
SELECT last_name FROM employees;
You can specify a statement ID when using the INTO clause, as in the following
statement:
EXPLAIN PLAN
SET STATEMENT_ID = 'st1'
INTO my_plan_table FOR
SELECT last_name FROM employees;
See Also:
Oracle Database SQL Language Reference for a complete description of
EXPLAIN PLAN syntax.
6-10
Chapter 6
Displaying PLAN_TABLE Output
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY('MY_PLAN_TABLE', 'st1','TYPICAL'));
See Also:
Oracle Database PL/SQL Packages and Types Reference for more
information about the DBMS_XPLAN package
-----------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
-----------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 3 | 189 | 10 (10)|
| 1 | NESTED LOOPS | | 3 | 189 | 10 (10)|
| 2 | NESTED LOOPS | | 3 | 141 | 7 (15)|
|* 3 | TABLE ACCESS FULL | EMPLOYEES | 3 | 60 | 4 (25)|
| 4 | TABLE ACCESS BY INDEX ROWID| JOBS | 19 | 513 | 2 (50)|
|* 5 | INDEX UNIQUE SCAN | JOB_ID_PK | 1 | | |
6-11
Chapter 6
Displaying PLAN_TABLE Output
-------------------------------------------------------------------------------------------
-----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
-------------------------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 3 | 189 | 8 (13)|
00:00:01 |
| 1 | NESTED LOOPS | | | |
| |
| 2 | NESTED LOOPS | | 3 | 189 | 8 (13)|
00:00:01 |
| 3 | MERGE JOIN | | 3 | 141 | 5 (20)|
00:00:01 |
| 4 | TABLE ACCESS BY INDEX ROWID | JOBS | 19 | 513 | 2 (0)|
00:00:01 |
| 5 | INDEX FULL SCAN | JOB_ID_PK | 19 | | 1 (0)|
00:00:01 |
|* 6 | SORT JOIN | | 3 | 60 | 3 (34)|
00:00:01 |
| 7 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES | 3 | 60 | 2 (0)|
00:00:01 |
|* 8 | INDEX RANGE SCAN | EMP_EMP_ID_PK | 3 | | 1 (0)|
00:00:01 |
|* 9 | INDEX UNIQUE SCAN | DEPT_ID_PK | 1 | | 0 (0)|
00:00:01 |
| 10 | TABLE ACCESS BY INDEX ROWID | DEPARTMENTS | 1 | 16 | 1 (0)|
00:00:01 |
-------------------------------------------------------------------------------------------
-----
6 - access("E"."JOB_ID"="J"."JOB_ID")
filter("E"."JOB_ID"="J"."JOB_ID")
8 - access("E"."EMPLOYEE_ID"<103)
9 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
6-12
Chapter 6
Displaying PLAN_TABLE Output
For example:
• Start with ID = 0 and given STATEMENT_ID.
• Use the CONNECT BY clause to walk the tree from parent to child, the join keys
being STATEMENT_ID = PRIOR STATMENT_ID and PARENT_ID = PRIOR ID.
• Use the pseudo-column LEVEL (associated with CONNECT BY) to indent the children.
Rows Plan
------- ----------------------------------------
SELECT STATEMENT
TABLE ACCESS FULL EMPLOYEES
The NULL in the Rows column indicates that the optimizer does not have any
statistics on the table. Analyzing the table shows the following:
Rows Plan
------- ----------------------------------------
16957 SELECT STATEMENT
16957 TABLE ACCESS FULL EMPLOYEES
You can also select the COST. This is useful for comparing execution plans or for
understanding why the optimizer chooses one execution plan over another.
Note:
These simplified examples are not valid for recursive SQL.
6-13
7
Reading Execution Plans
Execution plans are represented as a tree of operations.
This chapter contains the following topics:
• Reading Execution Plans: Basic
This section uses EXPLAIN PLAN examples to illustrate execution plans.
• Reading Execution Plans: Advanced
In some cases, execution plans can be complicated and challenging to read.
• Execution Plan Reference
This section describes V$ views and PLAN_COLUMN columns.
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY(NULL, 'statement_id','BASIC'));
Examples of the output from this statement are shown in Example 7-4 and
Example 7-1.
Example 7-1 EXPLAIN PLAN for Statement ID ex_plan1
The following plan shows execution of a SELECT statement. The table employees is
accessed using a full table scan. Every row in the table employees is accessed, and
the WHERE clause criteria is evaluated for every row.
EXPLAIN PLAN
SET statement_id = 'ex_plan1' FOR
SELECT phone_number
FROM employees
WHERE phone_number LIKE '650%';
---------------------------------------
| Id | Operation | Name |
---------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | TABLE ACCESS FULL| EMPLOYEES |
---------------------------------------
7-1
Chapter 7
Reading Execution Plans: Advanced
EXPLAIN PLAN
SET statement_id = 'ex_plan2' FOR
SELECT last_name
FROM employees
WHERE last_name LIKE 'Pe%';
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY(NULL, 'ex_plan2','BASIC'));
----------------------------------------
| Id | Operation | Name |
----------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | INDEX RANGE SCAN| EMP_NAME_IX |
----------------------------------------
7-2
Chapter 7
Reading Execution Plans: Advanced
Assumptions
This tutorial assumes the following:
• The STATISTICS_LEVEL initialization parameter is set to ALL.
• The database uses the default settings for adaptive execution.
• As user oe, you want to issue the following separate queries:
SELECT o.order_id, v.product_name
FROM orders o,
( SELECT order_id, product_name
FROM order_items o, product_information p
WHERE p.product_id = o.product_id
AND list_price < 50
AND min_price < 40 ) v
WHERE o.order_id = v.order_id
SELECT product_name
FROM order_items o, product_information p
WHERE o.unit_price = 15
AND quantity > 1
AND p.product_id = o.product_id
• Before executing each query, you want to query DBMS_XPLAN.DISPLAY_PLAN to see
the default plan, that is, the plan that the optimizer chose before applying its
adaptive mechanism.
• After executing each query, you want to query DBMS_XPLAN.DISPLAY_CURSOR to see
the final plan and adaptive query plan.
• SYS has granted oe the following privileges:
– GRANT SELECT ON V_$SESSION TO oe
– GRANT SELECT ON V_$SQL TO oe
– GRANT SELECT ON V_$SQL_PLAN TO oe
– GRANT SELECT ON V_$SQL_PLAN_STATISTICS_ALL TO oe
7-3
Chapter 7
Reading Execution Plans: Advanced
The following sample output has been reformatted to fit on the page. In this plan,
the optimizer chooses a nested loops join. The original optimizer estimates are
shown in the E-Rows column, whereas the actual statistics gathered during
execution are shown in the A-Rows column. In the MERGE JOIN operation, the
difference between the estimated and actual number of rows is significant.
--------------------------------------------------------------------------------------------
|Id| Operation | Name |Start|E-Rows|A-Rows|A-Time|Buff|OMem|1Mem|O/1/M|
--------------------------------------------------------------------------------------------
| 0| SELECT STATEMENT | | 1| | 269|00:00:00.09|1338| | | |
| 1| NESTED LOOPS | | 1| 1| 269|00:00:00.09|1338| | | |
| 2| MERGE JOIN CARTESIAN| | 1| 4|9135|00:00:00.03| 33| | | |
|*3| TABLE ACCESS FULL |PRODUCT_INFORMAT| 1| 1| 87|00:00:00.01| 32| | | |
| 4| BUFFER SORT | | 87|105|9135|00:00:00.01| 1|4096|4096|1/0/0|
| 5| INDEX FULL SCAN | ORDER_PK | 1|105| 105|00:00:00.01| 1| | | |
|*6| INDEX UNIQUE SCAN | ORDER_ITEMS_UK |9135| 1| 269|00:00:00.03|1305| | | |
--------------------------------------------------------------------------------------------
7-4
Chapter 7
Reading Execution Plans: Advanced
2 - access("P"."PRODUCT_ID"="O"."PRODUCT_ID")
3 - filter(("MIN_PRICE"<40 AND "LIST_PRICE"<50))
5 - access("O"."ORDER_ID"="ORDER_ID")
Note
-----
- statistics feedback used for this statement
6. Query V$SQL to verify the performance improvement.
The following query shows the performance of the two statements (sample output
included).
SELECT CHILD_NUMBER, CPU_TIME, ELAPSED_TIME, BUFFER_GETS
FROM V$SQL
WHERE SQL_ID = 'gm2npz344xqn8';
The second statement executed, which is child number 1, used statistics feedback.
CPU time, elapsed time, and buffer gets are all significantly lower.
7. Explain the plan for the query of order_items.
For example, use the following statement:
EXPLAIN PLAN FOR
SELECT product_name
FROM order_items o, product_information p
WHERE o.unit_price = 15
AND quantity > 1
AND p.product_id = o.product_id
8. View the plan in the plan table.
For example, run the following statement:
SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY);
7-5
Chapter 7
Reading Execution Plans: Advanced
Sample output appears below. Based on statistics collected at run time (Step 4),
the optimizer chose a hash join rather than the nested loops join. The dashes (-)
indicate the steps in the nested loops plan that the optimizer considered but do not
ultimately choose. The switch illustrates the adaptive query plan feature.
-------------------------------------------------------------------------------
|Id | Operation | Name |Rows|Bytes|Cost(%CPU)|Time |
-------------------------------------------------------------------------------
| 0| SELECT STATEMENT | |4|128|7(0)|00:00:01|
| *1| HASH JOIN | |4|128|7(0)|00:00:01|
|- 2| NESTED LOOPS | | | | | |
|- 3| NESTED LOOPS | | |128|7(0)|00:00:01|
|- 4| STATISTICS COLLECTOR | | | | | |
| *5| TABLE ACCESS FULL | ORDER_ITEMS |4| 48|3(0)|00:00:01|
|-*6| INDEX UNIQUE SCAN | PRODUCT_INFORMATI_PK|1| |0(0)|00:00:01|
|- 7| TABLE ACCESS BY INDEX ROWID| PRODUCT_INFORMATION |1| 20|1(0)|00:00:01|
| 8| TABLE ACCESS FULL | PRODUCT_INFORMATION |1| 20|1(0)|00:00:01|
-------------------------------------------------------------------------------
Note
-----
- this is an adaptive plan (rows marked '-' are inactive)
7-6
Chapter 7
Reading Execution Plans: Advanced
See Also:
TRANSACTION
ACCOUNT
CUSTOMER
In this example, customer is the smallest table, whereas transaction is the largest
table. A typical OLTP query retrieves transaction information about a specific customer
7-7
Chapter 7
Reading Execution Plans: Advanced
account. The query drives from the customer table. The goal is to minimize logical I/O,
which typically minimizes other critical resources including physical I/O and CPU time.
For parallel queries, the driving table is usually the largest table. It would not be
efficient to use parallel query in this case because only a few rows from each table are
accessed. However, what if it were necessary to identify all customers who had
transactions of a certain type last month? It would be more efficient to drive from the
transaction table because no limiting conditions exist on the customer table. The
database would join rows from the transaction table to the account table, and then
finally join the result set to the customer table. In this case, the used on the account
and customer table are probably highly selective primary key or unique indexes rather
than the non-unique indexes used in the first query. Because the transaction table is
large and the column is not selective, it would be beneficial to use parallel query
driving from the transaction table.
See Also:
The OTHER_TAG column in "PLAN_TABLE Columns"
7-8
Chapter 7
Reading Execution Plans: Advanced
into the parallel plan. This plan is the same plan for all parallel execution servers when
executed in parallel or for the QC when executed serially.
Example 7-3 Parallel Query Explain Plan
The following simple example illustrates an EXPLAIN PLAN for a parallel query:
---------------------------------------------------------------------------
---------------------
| Id | Operation | Name | Rows| Bytes |Cost %CPU| TQ
|IN-OUT| PQ Distrib |
---------------------------------------------------------------------------
---------------------
| 0 | SELECT STATEMENT | | 107 | 2782 | 3 (34) |
| | |
| 1 | PX COORDINATOR | | | | |
| | |
| 2 | PX SEND QC (RANDOM) | :TQ10001 | 107 | 2782 | 3 (34) | Q1,01
| P->S | QC (RAND) |
| 3 | HASH GROUP BY | | 107 | 2782 | 3 (34) | Q1,01
| PCWP | |
| 4 | PX RECEIVE | | 107 | 2782 | 3 (34) | Q1,01
| PCWP | |
| 5 | PX SEND HASH | :TQ10000 | 107 | 2782 | 3 (34) | Q1,00
| P->P | HASH |
| 6 | HASH GROUP BY | | 107 | 2782 | 3 (34) | Q1,00
| PCWP | |
| 7 | PX BLOCK ITERATOR | | 107 | 2782 | 2 (0) | Q1,00
| PCWP | |
| 8 | TABLE ACCESS FULL| EMP2 | 107 | 2782 | 2 (0) | Q1,00
| PCWP | |
---------------------------------------------------------------------------
---------------------
One set of parallel execution servers scans EMP2 in parallel, while the second set
performs the aggregation for the GROUP BY operation. The PX BLOCK ITERATOR row
source represents the splitting up of the table EMP2 into pieces to divide the scan
workload between the parallel execution servers. The PX SEND and PX RECEIVE row
sources represent the pipe that connects the two sets of parallel execution servers as
rows flow up from the parallel scan, get repartitioned through the HASH table queue,
and then read by and aggregated on the top set. The PX SEND QC row source
represents the aggregated values being sent to the QC in random (RAND) order. The
PX COORDINATOR row source represents the QC or Query Coordinator which controls
and schedules the parallel plan appearing below it in the plan tree.
7-9
Chapter 7
Reading Execution Plans: Advanced
Note:
Queries using bitmap join index indicate the bitmap join index access path.
The operation for bitmap join index is the same as bitmap index.
SELECT STATEMENT
TABLE ACCESS T BY INDEX ROWID
BITMAP CONVERSION TO ROWID
BITMAP OR
BITMAP MINUS
BITMAP MINUS
BITMAP INDEX C1_IND SINGLE VALUE
BITMAP INDEX C2_IND SINGLE VALUE
BITMAP INDEX C2_IND SINGLE VALUE
BITMAP MERGE
BITMAP INDEX C3_IND RANGE SCAN
7-10
Chapter 7
Reading Execution Plans: Advanced
To view the EXPLAIN PLAN for this query, use the following command:
The EXPLAIN PLAN output for this query should look similar to the following:
---------------------------------------------------------------------------
-----
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
---------------------------------------------------------------------------
-----
|0| SELECT STATEMENT | | 11 | 77 | 4 (25)|
00:00:01|
|1| RESULT CACHE |b06ppfz9pxzstbttpbqyqnfbmy| | |
| |
|2| HASH GROUP BY | | 11 | 77 | 4 (25)|
00:00:01|
|3| TABLE ACCESS FULL| EMP |107 | 749| 3 (0) |
00:00:01|
---------------------------------------------------------------------------
-----
In this EXPLAIN PLAN, the ResultCache operator is identified by its CacheId, which is
b06ppfz9pxzstbttpbqyqnfbmy. You can now run a query on the
V$RESULT_CACHE_OBJECTS view by using this CacheId.
A join is implemented using partial partition-wise join if the DISTRIBUTION column of the
plan table of one of the joined tables contains PARTITION(KEY). Partial partition-wise
join is possible if one of the joined tables is partitioned on its join column and the table
is parallelized.
A join is implemented using full partition-wise join if the partition row source appears
before the join row source in the EXPLAIN PLAN output. Full partition-wise joins are
possible only if both joined tables are equipartitioned on their respective join columns.
Examples of execution plans for several types of partitioning follow.
This section contains the following topics:
7-11
Chapter 7
Reading Execution Plans: Advanced
Assume that the tables employees and departments from the Oracle Database sample
schema exist.
--------------------------------------------------------------------
|Id| Operation | Name |Rows| Bytes|Cost|Pstart|Pstop|
7-12
Chapter 7
Reading Execution Plans: Advanced
--------------------------------------------------------------------
| 0| SELECT STATEMENT | | 105| 13965 | 2 | | |
| 1| PARTITION RANGE ALL| | 105| 13965 | 2 | 1 | 5 |
| 2| TABLE ACCESS FULL | EMP_RANGE | 105| 13965 | 2 | 1 | 5 |
--------------------------------------------------------------------
The database creates a partition row source on top of the table access row source. It
iterates over the set of partitions to be accessed. In this example, the partition iterator
covers all partitions (option ALL), because a predicate was not used for pruning. The
PARTITION_START and PARTITION_STOP columns of the PLAN_TABLE show access to all
partitions from 1 to 5.
For the next example, consider the following statement:
-----------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes|Cost|Pstart|Pstop|
-----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 3 | 399 | 2 | | |
| 1 | PARTITION RANGE ITERATOR| | 3 | 399 | 2 | 4 | 5 |
| *2 | TABLE ACCESS FULL |EMP_RANGE| 3 | 399 | 2 | 4 | 5 |
-----------------------------------------------------------------------
In the previous example, the partition row source iterates from partition 4 to 5 because
the database prunes the other partitions using a predicate on hire_date.
-----------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes|Cost|Pstart|Pstop|
-----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 133 | 2 | | |
| 1 | PARTITION RANGE SINGLE| | 1 | 133 | 2 | 1 | 1 |
|* 2 | TABLE ACCESS FULL | EMP_RANGE | 1 | 133 | 2 | 1 | 1 |
-----------------------------------------------------------------------
In the previous example, only partition 1 is accessed and known at compile time; thus,
there is no need for a partition row source.
7-13
Chapter 7
Reading Execution Plans: Advanced
Note:
Oracle Database displays the same information for hash partitioned objects,
except the partition row source name is PARTITION HASH instead of
PARTITION RANGE. Also, with hash partitioning, pruning is only possible using
equality or IN-list predicates.
-----------------------------------------------------------------------
|Id| Operation | Name | Rows | Bytes |Cost|Pstart|Pstop|
-----------------------------------------------------------------------
| 0| SELECT STATEMENT | | 10120 | 1314K| 78 | | |
| 1| PARTITION RANGE ALL| | 10120 | 1314K| 78 | 1 | 5 |
| 2| PARTITION HASH ALL| | 10120 | 1314K| 78 | 1 | 3 |
| 3| TABLE ACCESS FULL| EMP_COMP | 10120 | 1314K| 78 | 1 | 15 |
-----------------------------------------------------------------------
This example shows the plan when Oracle Database accesses all subpartitions of all
partitions of a composite object. The database uses two partition row sources for this
purpose: a range partition row source to iterate over the partitions, and a hash partition
row source to iterate over the subpartitions of each accessed partition.
In the following example, the range partition row source iterates from partition 1 to 5,
because the database performs no pruning. Within each partition, the hash partition
row source iterates over subpartitions 1 to 3 of the current partition. As a result, the
table access row source accesses subpartitions 1 to 15. In other words, the database
accesses all subpartitions of the composite object.
7-14
Chapter 7
Reading Execution Plans: Advanced
FROM emp_comp
WHERE hire_date = TO_DATE('15-FEB-1998', 'DD-MON-YYYY');
-----------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes |Cost|Pstart|Pstop|
-----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 20 | 2660 | 17 | | |
| 1 | PARTITION RANGE SINGLE| | 20 | 2660 | 17 | 5 | 5 |
| 2 | PARTITION HASH ALL | | 20 | 2660 | 17 | 1 | 3 |
|* 3 | TABLE ACCESS FULL | EMP_COMP | 20 | 2660 | 17 | 13 | 15 |
-----------------------------------------------------------------------
In the previous example, only the last partition, partition 5, is accessed. This partition is
known at compile time, so the database does not need to show it in the plan. The hash
partition row source shows accessing of all subpartitions within that partition; that is,
subpartitions 1 to 3, which translates into subpartitions 13 to 15 of the emp_comp table.
------------------------------------------------------------------------
| Id | Operation |Name |Rows | Bytes |Cost|Pstart|Pstop|
------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 101 | 13433 | 78 | | |
| 1 | PARTITION RANGE ALL | | 101 | 13433 | 78 | 1 | 5 |
| 2 | PARTITION HASH SINGLE| | 101 | 13433 | 78 | 3 | 3 |
|* 3 | TABLE ACCESS FULL | EMP_COMP | 101 | 13433 | 78 | | |
------------------------------------------------------------------------
In the previous example, the predicate deptno=20 enables pruning on the hash
dimension within each partition. Therefore, Oracle Database only needs to access a
single subpartition. The number of this subpartition is known at compile time, so the
hash partition row source is not needed.
Finally, consider the following statement:
-----------------------------------------------------------------------
| Id| Operation | Name |Rows| Bytes |Cost|Pstart|Pstop|
-----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 101| 13433 | 78 | | |
| 1 | PARTITION RANGE ALL | | 101| 13433 | 78 | 1 | 5 |
| 2 | PARTITION HASH SINGLE| | 101| 13433 | 78 | KEY | KEY |
|*3 | TABLE ACCESS FULL | EMP_COMP | 101| 13433 | 78 | | |
-----------------------------------------------------------------------
7-15
Chapter 7
Reading Execution Plans: Advanced
The last two examples are the same, except that department_id = :dno replaces
deptno=20. In this last case, the subpartition number is unknown at compile time, and
a hash partition row source is allocated. The option is SINGLE for this row source
because Oracle Database accesses only one subpartition within each partition. In Step
2, both PARTITION_START and PARTITION_STOP are set to KEY. This value means that
Oracle Database determines the number of subpartitions at run time.
-------------------------------------------------------------------------------------------
-----
|Id| Operation |Name |Row|Byte|Cost|Pstart|Pstop|TQ|IN-OUT|PQ
Distrib|
-------------------------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | |284 |16188|6 | | | |
| |
| 1| PX COORDINATOR | | | | | | | |
| |
| 2| PX SEND QC (RANDOM) |:TQ10001 |284 |16188|6 | | | Q1,01 |P->S |QC
(RAND) |
|*3| HASH JOIN | |284 |16188|6 | | | Q1,01 |PCWP
| |
| 4| PX PARTITION RANGE ALL | |284 |7668 |2 | 1 | 2 | Q1,01 |PCWC
| |
| 5| TABLE ACCESS FULL |EMP_RANGE_DID|284 |7668 |2 | 1 | 2 | Q1,01 |PCWP
| |
| 6| BUFFER SORT | | | | | | | Q1,01 |PCWC
| |
| 7| PX RECEIVE | | 21 | 630 |2 | | | Q1,01 |PCWP
| |
7-16
Chapter 7
Reading Execution Plans: Advanced
The execution plan shows that the table dept2 is scanned serially and all rows with the
same partitioning column value of emp_range_did (department_id) are sent through
a PART (KEY), or partition key, table queue to the same parallel execution server doing
the partial partition-wise join.
Example 7-6 Partial Partition-Wise Join with Composite Partition
In the following example, emp_comp is joined on the partitioning column and is
parallelized, enabling use of a partial partition-wise join because dept2 is not
partitioned. The database dynamically partitions dept2 before the join.
-------------------------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows |Bytes |Cost|Pstart|Pstop|TQ |IN-OUT|PQ
Distrib|
-------------------------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 445 | 17800 | 5 | | | |
| |
| 1 | PX COORDINATOR | | | | | | | |
| |
| 2 | PX SEND QC (RANDOM) |:TQ10001| 445 | 17800 | 5 | | | Q1,01 | P->S | QC
(RAND)|
|*3 | HASH JOIN | | 445 | 17800 | 5 | | | Q1,01 | PCWP
| |
| 4 | PX PARTITION RANGE ALL | | 107 | 1070 | 3 | 1 | 5 | Q1,01 | PCWC
| |
| 5 | PX PARTITION HASH ALL | | 107 | 1070 | 3 | 1 | 3 | Q1,01 | PCWC
| |
| 6 | TABLE ACCESS FULL |EMP_COMP| 107 | 1070 | 3 | 1 | 15| Q1,01 | PCWP
| |
| 7 | PX RECEIVE | | 21 | 630 | 1 | | | Q1,01 | PCWP
| |
| 8 | PX SEND PARTITION (KEY)|:TQ10000| 21 | 630 | 1 | | | Q1,00 | P->P |PART
(KEY)|
| 9 | PX BLOCK ITERATOR | | 21 | 630 | 1 | | | Q1,00 | PCWC
| |
|10 | TABLE ACCESS FULL |DEPT2 | 21 | 630 | 1 | | | Q1,00 | PCWP
7-17
Chapter 7
Reading Execution Plans: Advanced
| |
-------------------------------------------------------------------------------------------
-----
The plan shows that the optimizer selects partial partition-wise join from one of two
columns. The PX SEND node type is PARTITION (KEY) and the PQ Distrib column
contains the text PART (KEY), or partition key. This implies that the table dept2 is re-
partitioned based on the join column department_id to be sent to the parallel
execution servers executing the scan of EMP_COMP and the join.
---------------------------------------------------------------------------
---------------------
|Id| Operation | Name |Rows|Bytes|Cost|Pstart|Pstop|
TQ |IN-OUT|PQ Distrib|
---------------------------------------------------------------------------
---------------------
| 0| SELECT STATEMENT | | 106 | 2544 | 8 | |
| | | |
| 1| PX COORDINATOR | | | | | |
| | | |
| 2| PX SEND QC (RANDOM) | :TQ10000 | 106 | 2544 | 8 | | |
Q1,00 | P->S |QC (RAND)|
| 3| PX PARTITION HASH ALL | | 106 | 2544 | 8 | 1 | 3 |
Q1,00 | PCWC | |
|*4| HASH JOIN | | 106 | 2544 | 8 | | |
Q1,00 | PCWP | |
| 5| PX PARTITION RANGE ALL| | 107 | 1070 | 3 | 1 | 5 |
Q1,00 | PCWC | |
| 6| TABLE ACCESS FULL | EMP_COMP | 107 | 1070 | 3 | 1 | 15 |
Q1,00 | PCWP | |
| 7| TABLE ACCESS FULL | DEPT_HASH | 27 | 378 | 4 | 1 | 3 |
Q1,00 | PCWP | |
---------------------------------------------------------------------------
---------------------
7-18
Chapter 7
Reading Execution Plans: Advanced
The PX PARTITION HASH row source appears on top of the join row source in the plan
table output while the PX PARTITION RANGE row source appears over the scan of
emp_comp. Each parallel execution server performs the join of an entire hash partition
of emp_comp with an entire partition of dept_hash.
The INLIST ITERATOR operation iterates over the next operation in the plan for each
value in the IN-list predicate. The following sections describe the three possible types
of IN-list columns for partitioned tables and indexes.
7-19
Chapter 7
Reading Execution Plans: Advanced
INLIST ITERATOR
TABLE ACCESS BY LOCAL INDEX ROWID EMP KEY(INLIST)
KEY(INLIST)
INDEX RANGE SCAN EMP_EMPNO KEY(INLIST)
KEY(INLIST)
The KEY(INLIST) designation for the partition start and stop keys specifies that an IN-
list predicate appears on the index start and stop keys.
7.2.5.5.2 When the IN-List Column is an Index and a Partition Column: Example
If empno is an indexed and a partition column, then the plan contains an INLIST
ITERATOR operation before the partition operation.
7-20
Chapter 7
Reading Execution Plans: Advanced
7-21
Chapter 7
Reading Execution Plans: Advanced
7-22
Chapter 7
Reading Execution Plans: Advanced
7-23
Chapter 7
Reading Execution Plans: Advanced
7-24
Chapter 7
Reading Execution Plans: Advanced
Table 7-2 describes the values that can appear in the DISTRIBUTION column:
PARTITION (KEY) Maps rows to query servers based on the partitioning of a table or index using a set of
columns. Used for partial partition-wise join, PARALLEL INSERT, CREATE TABLE AS
SELECT of a partitioned table, and CREATE PARTITIONED GLOBAL INDEX.
HASH Maps rows to query servers using a hash function on the join key. Used for PARALLEL
JOIN or PARALLEL GROUP BY.
7-25
Chapter 7
Reading Execution Plans: Advanced
Table 7-3 lists each combination of OPERATION and OPTIONS produced by the EXPLAIN
PLAN statement and its meaning within an execution plan.
7-26
Chapter 7
Reading Execution Plans: Advanced
7-27
Chapter 7
Reading Execution Plans: Advanced
INDEX FULL SCAN Retrieval of all rowids from an index when there is no
start or stop key. Indexed values are scanned in
ascending order.
INDEX FULL SCAN Retrieval of all rowids from an index when there is no
DESCENDING start or stop key. Indexed values are scanned in
descending order.
INDEX FAST FULL Retrieval of all rowids (and column values) using
SCAN multiblock reads. No sorting order can be defined.
Compares to a full table scan on only the indexed
columns. Only available with the cost based optimizer.
INDEX SKIP SCAN Retrieval of rowids from a concatenated index without
using the leading column(s) in the index. Only available
with the cost based optimizer.
INLIST Iterates over the next operation in the plan for each
ITERATOR value in the IN-list predicate.
7-28
Chapter 7
Reading Execution Plans: Advanced
MAT_VIEW BY USER ROWID If the materialized view rows are located using user-
REWITE ACCESS supplied rowids.
7-29
Chapter 7
Reading Execution Plans: Advanced
7-30
Chapter 7
Reading Execution Plans: Advanced
TABLE ACCESS BY USER ROWID If the table rows are located using user-supplied
rowids.
TABLE ACCESS BY INDEX If the table is nonpartitioned and rows are located using
ROWID indexes.
TABLE ACCESS BY GLOBAL If the table is partitioned and rows are located using
INDEX ROWID only global indexes.
TABLE ACCESS BY LOCAL If the table is partitioned and rows are located using
INDEX ROWID one or more local indexes and possibly some global
indexes.
Partition Boundaries:
The partition boundaries might have been computed
by:
A previous PARTITION step, in which case the
PARTITION_START and PARTITION_STOP column
values replicate the values present in the PARTITION
step, and the PARTITION_ID contains the ID of the
PARTITION step. Possible values for
PARTITION_START and PARTITION_STOP are
NUMBER(n), KEY, INVALID.
The TABLE ACCESS or INDEX step itself, in which case
the PARTITION_ID contains the ID of the step.
Possible values for PARTITION_START and
PARTITION_STOP are NUMBER(n), KEY, ROW
REMOVE_LOCATION (TABLE ACCESS only), and
INVALID.
7-31
Chapter 7
Execution Plan Reference
See Also:
Oracle Database Reference for more information about PLAN_TABLE
View Description
V$SQL_SHARED_CURSOR Explains why a particular child cursor is not shared with
existing child cursors. Each column identifies a specific
reason why the cursor cannot be shared.
The USE_FEEDBACK_STATS column shows whether a
child cursor fails to match because of reoptimization.
7-32
Chapter 7
Execution Plan Reference
View Description
V$SQL_PLAN Includes a superset of all rows appearing in all final
plans. PLAN_LINE_ID is consecutively numbered, but
for a single final plan, the IDs may not be consecutive.
V$SQL_PLAN_STATISTICS_ALL Contains memory usage statistics for row sources that
use SQL memory (sort or hash join). This view
concatenates information in V$SQL_PLAN with
execution statistics from V$SQL_PLAN_STATISTICS
and V$SQL_WORKAREA.
TIMESTAMP DATE Date and time when the EXPLAIN PLAN statement was
generated.
REMARKS VARCHAR2(80) Any comment (of up to 80 bytes) you want to associate
with each step of the explained plan. This column
indicates whether the database used an outline or SQL
profile for the query.
If you need to add or change a remark on any row of the
PLAN_TABLE, then use the UPDATE statement to modify
the rows of the PLAN_TABLE.
OPERATION VARCHAR2(30) Name of the internal operation performed in this step. In
the first row generated for a statement, the column
contains one of the following values:
• DELETE STATEMENT
• INSERT STATEMENT
• SELECT STATEMENT
• UPDATE STATEMENT
See Table 7-6 for more information about values for this
column.
OPTIONS VARCHAR2(225) A variation on the operation that the OPERATION column
describes.
See Table 7-6 for more information about values for this
column.
7-33
Chapter 7
Execution Plan Reference
PARENT_ID NUMERIC The ID of the next execution step that operates on the
output of the ID step.
DEPTH NUMERIC Depth of the operation in the row source tree that the
plan represents. You can use this value to indent the
rows in a plan table report.
POSITION NUMERIC For the first row of output, this indicates the optimizer's
estimated cost of executing the statement. For the other
rows, it indicates the position relative to the other
children of the same parent.
COST NUMERIC Cost of the operation as estimated by the optimizer's
query approach. Cost is not determined for table access
operations. The value of this column does not have any
particular unit of measurement; it is a weighted value
used to compare costs of execution plans. The value of
this column is a function of the CPU_COST and IO_COST
columns.
CARDINALITY NUMERIC Estimate by the query optimization approach of the
number of rows that the operation accessed.
BYTES NUMERIC Estimate by the query optimization approach of the
number of bytes that the operation accessed.
7-34
Chapter 7
Execution Plan Reference
7-35
Chapter 7
Execution Plan Reference
7-36
Chapter 7
Execution Plan Reference
BROADCAST Broadcasts the rows of the entire table to each query server. Used for a parallel
join when one table is very small compared to the other.
QC (ORDER) The QC consumes the input in order, from the first to the last query server.
Used when the statement contains an ORDER BY clause.
QC (RANDOM) The QC consumes the input randomly. Used when the statement does not have
an ORDER BY clause.
Table 7-7 lists each combination of OPERATION and OPTIONS produced by the EXPLAIN
PLAN statement and its meaning within an execution plan.
7-37
Chapter 7
Execution Plan Reference
Table 7-7 (Cont.) OPERATION and OPTIONS Values Produced by EXPLAIN PLAN
7-38
Chapter 7
Execution Plan Reference
Table 7-7 (Cont.) OPERATION and OPTIONS Values Produced by EXPLAIN PLAN
INDEX FULL SCAN Retrieval of all rowids from an index when there is no start or stop
key. Indexed values are scanned in ascending order.
INDEX FULL SCAN Retrieval of all rowids from an index when there is no start or stop
DESCENDING key. Indexed values are scanned in descending order.
INDEX FAST FULL SCAN Retrieval of all rowids (and column values) using multiblock reads.
No sorting order can be defined. Compares to a full table scan on
only the indexed columns. Only available with the cost based
optimizer.
INDEX SKIP SCAN Retrieval of rowids from a concatenated index without using the
leading column(s) in the index. Only available with the cost based
optimizer.
INLIST ITERATOR Iterates over the next operation in the plan for each value in the IN-
list predicate.
INTERSECTION Operation accepting two sets of rows and returning the intersection
of the sets, eliminating duplicates.
MERGE JOIN Operation accepting two sets of rows, each sorted by a value,
(These are join combining each row from one set with the matching rows from the
operations.) other, and returning the result.
MERGE JOIN OUTER Merge join operation to perform an outer join statement.
MERGE JOIN ANTI Merge antijoin.
MERGE JOIN SEMI Merge semijoin.
MERGE JOIN CARTESIAN Can result from 1 or more of the tables not having any join
conditions to any other tables in the statement. Can occur even with
a join and it may not be flagged as CARTESIAN in the plan.
7-39
Chapter 7
Execution Plan Reference
Table 7-7 (Cont.) OPERATION and OPTIONS Values Produced by EXPLAIN PLAN
MAT_VIEW REWITE HASH Retrieval of rows from materialized view based on hash cluster key
ACCESS value.
MAT_VIEW REWITE BY ROWID RANGE Retrieval of rows from a materialized view based on a rowid range.
ACCESS
MAT_VIEW REWITE SAMPLE BY ROWID Retrieval of sampled rows from a materialized view based on a
ACCESS RANGE rowid range.
MAT_VIEW REWITE BY USER ROWID If the materialized view rows are located using user-supplied rowids.
ACCESS
MAT_VIEW REWITE BY INDEX ROWID If the materialized view is nonpartitioned and rows are located using
ACCESS indexes.
MAT_VIEW REWITE BY GLOBAL INDEX If the materialized view is partitioned and rows are located using
ACCESS ROWID only global indexes.
MAT_VIEW REWITE BY LOCAL INDEX If the materialized view is partitioned and rows are located using one
ACCESS ROWID or more local indexes and possibly some global indexes.
Partition Boundaries:
The partition boundaries might have been computed by:
A previous PARTITION step, in which case the PARTITION_START
and PARTITION_STOP column values replicate the values present in
the PARTITION step, and the PARTITION_ID contains the ID of the
PARTITION step. Possible values for PARTITION_START and
PARTITION_STOP are NUMBER(n), KEY, and INVALID.
The MAT_VIEW REWRITE ACCESS or INDEX step itself, in which
case the PARTITION_ID contains the ID of the step. Possible
values for PARTITION_START and PARTITION_STOP are
NUMBER(n), KEY, ROW REMOVE_LOCATION (MAT_VIEW REWRITE
ACCESS only), and INVALID.
MINUS Operation accepting two sets of rows and returning rows appearing
in the first set but not in the second, eliminating duplicates.
NESTED LOOPS Operation accepting two sets of rows, an outer set and an inner set.
(These are join Oracle Database compares each row of the outer set with each row
operations.) of the inner set, returning rows that satisfy a condition. This join
method is useful for joining small subsets of data (OLTP). The join
condition is an efficient way of accessing the second table.
NESTED LOOPS OUTER Nested loops operation to perform an outer join statement.
7-40
Chapter 7
Execution Plan Reference
Table 7-7 (Cont.) OPERATION and OPTIONS Values Produced by EXPLAIN PLAN
7-41
Chapter 7
Execution Plan Reference
Table 7-7 (Cont.) OPERATION and OPTIONS Values Produced by EXPLAIN PLAN
TABLE ACCESS BY LOCAL INDEX If the table is partitioned and rows are located using one or more
ROWID local indexes and possibly some global indexes.
Partition Boundaries:
The partition boundaries might have been computed by:
A previous PARTITION step, in which case the PARTITION_START
and PARTITION_STOP column values replicate the values present in
the PARTITION step, and the PARTITION_ID contains the ID of the
PARTITION step. Possible values for PARTITION_START and
PARTITION_STOP are NUMBER(n), KEY, and INVALID.
The TABLE ACCESS or INDEX step itself, in which case the
PARTITION_ID contains the ID of the step. Possible values for
PARTITION_START and PARTITION_STOP are NUMBER(n), KEY, ROW
REMOVE_LOCATION (TABLE ACCESS only), and INVALID.
TRANSPOSE Operation evaluating a PIVOT operation by transposing the results
of GROUP BY to produce the final pivoted data.
UNION Operation accepting two sets of rows and returns the union of the
sets, eliminating duplicates.
UNPIVOT Operation that rotates data from columns into rows.
VIEW Operation performing a view's query and then returning the resulting
rows to another operation.
See Also:
Oracle Database Reference for more information about PLAN_TABLE
7-42
Chapter 7
Execution Plan Reference
The display functions accept options for displaying the plan table output. You can
specify:
• A plan table name if you are using a table different from PLAN_TABLE
• A statement ID if you have set a statement ID with the EXPLAIN PLAN
• A format option that determines the level of detail: BASIC, SERIAL, TYPICAL, ALL,
and in some cases ADAPTIVE
7-43
Chapter 7
Execution Plan Reference
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_XPLAN display functions
7-44
Part IV
SQL Operators: Access Paths and Joins
A row source is a set of rows returned by a step in the execution plan. A SQL
operator acts on a row source.
A unary operator acts on one input, as with access paths. A binary operator acts on
two outputs, as with joins.
This part contains the following chapters:
• Optimizer Access Paths
An access path is a technique used by a query to retrieve rows from a row
source.
• Joins
Oracle Database provides several optimizations for joining row sets.
8
Optimizer Access Paths
An access path is a technique used by a query to retrieve rows from a row source.
This chapter contains the following topics:
• Introduction to Access Paths
A row source is a set of rows returned by a step in an execution plan. A row
source can be a table, view, or result of a join or grouping operation.
• Table Access Paths
A table is the basic unit of data organization in an Oracle database.
• B-Tree Index Access Paths
An index is an optional structure, associated with a table or table cluster, that can
sometimes speed data access.
• Bitmap Index Access Paths
Bitmap indexes combine the indexed data with a rowid range.
• Table Cluster Access Paths
A table cluster is a group of tables that share common columns and store related
data in the same blocks. When tables are clustered, a single data block can
contain rows from multiple tables.
8-1
Chapter 8
Table Access Paths
The optimizer considers different possible execution plans, and then assigns each
plan a cost. The optimizer chooses the plan with the lowest cost. In general, index
access paths are more efficient for statements that retrieve a small subset of table
rows, whereas full table scans are more efficient when accessing a large portion of a
table.
See Also:
• "Joins"
• "Cost-Based Optimization"
• Oracle Database Concepts for an overview of these structures
8-2
Chapter 8
Table Access Paths
See Also:
8-3
Chapter 8
Table Access Paths
allocates one or more extents to form the data segment for a table. The database also
allocates one or more extents to form the index segment for a table.
By default, the database uses automatic segment space management (ASSM) for
permanent, locally managed tablespaces. When a session first inserts data into a
table, the database formats a bitmap block. The bitmap tracks the blocks in the
segment. The database uses the bitmap to find free blocks and then formats each
block before writing to it. ASSM spread out inserts among blocks to avoid concurrency
issues.
The high water mark (HWM) is the point in a segment beyond which data blocks are
unformatted and have never been used. Below the HWM, a block may be formatted
and written to, formatted and empty, or unformatted. The low high water mark (low
HWM) marks the point below which all blocks are known to be formatted because they
either contain data or formerly contained data.
During a full table scan, the database reads all blocks up to the low HWM, which are
known to be formatted, and then reads the segment bitmap to determine which blocks
between the HWM and low HWM are formatted and safe to read. The database knows
not to read past the HWM because these blocks are unformatted.
See Also:
Oracle Database Concepts to learn about data block storage
See Also:
Oracle Database Concepts to learn about rowids
8-4
Chapter 8
Table Access Paths
Direct path
read
Situations in which Oracle Database may perform direct path reads include:
• Execution of a CREATE TABLE AS SELECT statement
• Execution of an ALTER REBUILD or ALTER MOVE statement
• Reads from a temporary tablespace
• Parallel queries
• Reads from a LOB segment
See Also:
Oracle Database Performance Tuning Guide to learn about wait events for
direct path reads
8-5
Chapter 8
Table Access Paths
8-6
Chapter 8
Table Access Paths
8-7
Chapter 8
Table Access Paths
The following graphic depicts a scan of a table segment, showing how the scan skips
unformatted blocks below the high water mark.
Because the blocks are adjacent, the database can speed up the scan by making I/O
calls larger than a single block, known as a multiblock read. The size of a read call
ranges from one block to the number of blocks specified by the
DB_FILE_MULTIBLOCK_READ_COUNT initialization parameter. For example, setting this
parameter to 4 instructs the database to read up to 4 blocks in a single call.
The algorithms for caching blocks during full table scans are complex. For example,
the database caches blocks differently depending on whether tables are small or large.
See Also:
• "Table 19-2"
• Oracle Database Concepts for an overview of the default caching mode
• Oracle Database Reference to learn about the
DB_FILE_MULTIBLOCK_READ_COUNT initialization parameter
SELECT salary
FROM hr.employees
WHERE salary > 4000;
8-8
Chapter 8
Table Access Paths
---------------------------------------------------------------------------
----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT | | | | 3
(100)| |
|* 1 | TABLE ACCESS FULL| EMPLOYEES | 98 | 6762 | 3 (0)|
00:00:01 |
---------------------------------------------------------------------------
----
1 - filter("SALARY">4000)
Note:
Rowids can change between versions. Accessing data based on position is
not recommended because rows can move.
8-9
Chapter 8
Table Access Paths
See Also:
Oracle Database Development Guide to learn more about rowids
SELECT *
FROM employees
WHERE employee_id > 190;
Step 2 of the following plan shows a range scan of the emp_emp_id_pk index on the
hr.employees table. The database uses the rowids obtained from the index to find the
corresponding rows from the employees table, and then retrieve them. The BATCHED
access shown in Step 1 means that the database retrieves a few rowids from the
8-10
Chapter 8
Table Access Paths
index, and then attempts to access rows in block order to improve the clustering and
reduce the number of times that the database must access a block.
---------------------------------------------------------------------------
-----
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time|
---------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | | | |
2(100)| |
| 1| TABLE ACCESS BY INDEX ROWID BATCHED|EMPLOYEES |16|1104|2 (0)|
00:00:01|
|*2| INDEX RANGE SCAN |EMP_EMP_ID_PK|16| |1 (0)|
00:00:01|
---------------------------------------------------------------------------
-----
2 - access("EMPLOYEE_ID">190)
• SAMPLE (sample_percent)
The database reads a specified percentage of rows in the table to perform a
sample table scan.
• SAMPLE BLOCK (sample_percent)
The database reads a specified percentage of table blocks to perform a sample
table scan.
The sample_percent specifies the percentage of the total row or block count to include
in the sample. The value must be in the range .000001 up to, but not including, 100.
8-11
Chapter 8
Table Access Paths
This percentage indicates the probability of each row, or each cluster of rows in block
sampling, being selected for the sample. It does not mean that the database retrieves
exactly sample_percent of the rows.
Note:
Block sampling is possible only during full table scans or index fast full
scans. If a more efficient execution path exists, then the database does not
sample blocks. To guarantee block sampling for a specific table or index, use
the FULL or INDEX_FFS hint.
See Also:
The EXPLAIN PLAN output for this statement might look as follows:
-------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
-------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 68 | 3 (34)|
| 1 | TABLE ACCESS SAMPLE | EMPLOYEES | 1 | 68 | 3 (34)|
-------------------------------------------------------------------------
8-12
Chapter 8
Table Access Paths
See Also:
Oracle Database In-Memory Guide for an introduction to the IM column store
8-13
Chapter 8
Table Access Paths
See Also:
SELECT *
FROM oe.product_information
WHERE list_price > 10
ORDER BY product_id
The plan for this statement might look as follows, with the INMEMORY keyword in Step 2
indicating that some or all of the object was accessed from the IM column store:
8-14
Chapter 8
B-Tree Index Access Paths
2 - inmemory("LIST_PRICE">10)
filter("LIST_PRICE">10)
See Also:
8-15
Chapter 8
B-Tree Index Access Paths
A B-tree index is an ordered list of values divided into ranges. By associating a key
with a row or range of rows, B-trees provide excellent retrieval performance for a wide
range of queries, including exact match and range searches.
This section contains the following topics:
• B-Tree Index Structure
A B-tree index has two types of blocks: branch blocks for searching and leaf
blocks that store values.
• How Index Storage Affects Index Scans
Bitmap index blocks can appear anywhere in the index segment.
• Unique and Nonunique Indexes
In a nonunique index, the database stores the rowid by appending it to the key as
an extra column. The entry adds a length byte to make the key unique.
• B-Tree Indexes and Nulls
B-tree indexes never store completely null keys, which is important for how the
optimizer chooses access paths. A consequence of this rule is that single-column
B-tree indexes never store nulls.
8-16
Chapter 8
B-Tree Index Access Paths
Branch Blocks
0..40
41..80
81..120
....
200..250
Leaf Blocks
8-17
Chapter 8
B-Tree Index Access Paths
See Also:
Oracle Database Concepts to learn about index blocks
0,AAAPvCAAFAAAAFaAAa
0,AAAPvCAAFAAAAFaAAg
0,AAAPvCAAFAAAAFaAAl
2,AAAPvCAAFAAAAFaAAm
In a unique index, the index key does not include the rowid. The database sorts the
data only by the index key values, such as 0, 1, 2, and so on.
See Also:
Oracle Database Concepts for an overview of unique and nonunique indexes
COUNT(*)
----------
1
COUNT(*)
----------
0
8-18
Chapter 8
B-Tree Index Access Paths
The following example shows that the optimizer chooses a full table scan for a query
of all department IDs in hr.employees. The optimizer cannot use the index on
employees.department_id because the index is not guaranteed to include entries for
every row in the table.
Explained.
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Plan hash value: 3476115102
---------------------------------------------------------------------------
----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT | | 107 | 321 | 2 (0)|
00:00:01 |
| 1 | TABLE ACCESS FULL| EMPLOYEES | 107 | 321 | 2 (0)|
00:00:01 |
---------------------------------------------------------------------------
----
8 rows selected.
The following example shows the optimizer can use the index on department_id for a
query of a specific department ID because all non-null rows are indexed.
Explained.
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Plan hash value: 67425611
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows |Bytes| Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
8-19
Chapter 8
B-Tree Index Access Paths
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 1 | 3 | 1 (0)| 00:0
0:01|
|*1 | INDEX RANGE SCAN| EMP_DEPARTMENT_IX | 1 | 3 | 1 (0)| 00:0
0:01|
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
1 - access("DEPARTMENT_ID"=10)
The following example shows that the optimizer chooses an index scan when the
predicate excludes null values:
Explained.
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Plan hash value: 1590637672
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows|Bytes| Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 106 | 318 | 1 (0)| 00:0
0:01 |
|*1 | INDEX FULL SCAN | EMP_DEPARTMENT_IX | 106 | 318 | 1 (0)| 00:0
0:01 |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
8-20
Chapter 8
B-Tree Index Access Paths
UNIQUENES
---------
NONUNIQUE
The following code creates a primary key constraint on a column with a non-unique
index, resulting in an index range scan rather than an index unique scan:
Execution Plan
----------------------------------------------------------
Plan hash value: 868081059
---------------------------------------------------------------------------
-----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
8-21
Chapter 8
B-Tree Index Access Paths
-----
| 0 | SELECT STATEMENT | | 1 | 13 | 1 (0)|
00:00:01 |
|* 1 | INDEX RANGE SCAN| T_TABLE_IDX | 1 | 13 | 1 (0)|
00:00:01 |
---------------------------------------------------------------------------
-----
You can use the INDEX(alias index_name) hint to specify the index to use, but not a
specific type of index access path.
See Also:
• Oracle Database Concepts for more details on index structures and for
detailed information on how a B-tree is searched
• Oracle Database SQL Language Reference to learn more about the
INDEX hint
8-22
Chapter 8
B-Tree Index Access Paths
Branch Blocks
0..40
41..80
81..120
....
200..250
Leaf Blocks
The following statement queries the record for product 19 in the sh.products table:
SELECT *
FROM sh.products
WHERE prod_id = 19;
Because a primary key index exists on the products.prod_id column, and the WHERE
clause references all of the columns using an equality operator, the optimizer chooses
a unique scan:
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes|Cost (%CPU)|
8-23
Chapter 8
B-Tree Index Access Paths
Time |
---------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | | | | 1
(100)| |
| 1| TABLE ACCESS BY INDEX ROWID| PRODUCTS | 1 | 173 | 1 (0)|
00:00:01|
|* 2| INDEX UNIQUE SCAN | PRODUCTS_PK | 1 | | 0
(0)| |
---------------------------------------------------------------------------
-----
2 - access("PROD_ID"=19)
8-24
Chapter 8
B-Tree Index Access Paths
Note:
For the optimizer to consider a range scan, wild-card searches of the
form col1 LIKE '%ASD' must not be in a leading position.
Tip:
If you require sorted data, then use the ORDER BY clause, and do not rely on
an index. If an index can satisfy an ORDER BY clause, then the optimizer uses
this option and thereby avoids a sort.
The optimizer considers an index range scan descending when an index can satisfy
an ORDER BY DESCENDING clause.
If the optimizer chooses a full table scan or another index, then a hint may be required
to force this access path. The INDEX(tbl_alias ix_name) and INDEX_DESC(tbl_alias
ix_name) hints instruct the optimizer to use a specific index.
See Also:
Oracle Database SQL Language Reference to learn more about the INDEX
and INDEX_DESC hints
8-25
Chapter 8
B-Tree Index Access Paths
Note:
In some cases, an index scan reads a set of index blocks, sorts the
rowids, and then reads a set of table blocks.
Thus, to scan the index, the database moves backward or forward through the leaf
blocks. For example, a scan for IDs between 20 and 40 locates the first leaf block that
has the lowest key value that is 20 or greater. The scan proceeds horizontally through
the linked list of leaf nodes until it finds a value greater than 40, and then stops.
The following figure illustrates an index range scan using ascending order. A
statement requests the employees records with the value 20 in the department_id
column, which has a nonunique index. In this example, 2 index entries for department
20 exist.
Branch Blocks
0..40
41..80
81..120
....
200..250
Leaf Blocks
8-26
Chapter 8
B-Tree Index Access Paths
SELECT *
FROM employees
WHERE department_id = 20
AND salary > 1000;
The preceding query has low cardinality (returns few rows), so the query uses the
index on the department_id column. The database scans the index, fetches the
records from the employees table, and then applies the salary > 1000 filter to these
fetched records to generate the result.
-------------------------------------------------------------------------------------------
|Id | Operation | Name |Rows|Bytes|Cost(%CPU)| Time |
-------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 2 (100)| |
|*1 | TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES | 2 | 138 | 2 (0)|00:00:01|
|*2 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX| 2 | | 1 (0)|00:00:01|
-------------------------------------------------------------------------------------------
1 - filter("SALARY">1000)
2 - access("DEPARTMENT_ID"=20)
SELECT *
FROM employees
WHERE department_id < 20
ORDER BY department_id DESC;
8-27
Chapter 8
B-Tree Index Access Paths
This preceding query has low cardinality, so the query uses the index on the
department_id column.
2 - access("DEPARTMENT_ID"<20)
The database locates the first index leaf block that contains the highest key value that
is 20 or less. The scan then proceeds horizontally to the left through the linked list of
leaf nodes. The database obtains the rowid from each index entry, and then retrieves
the row specified by the rowid.
8-28
Chapter 8
B-Tree Index Access Paths
• A predicate references a column in the index. This column need not be the leading
column.
• No predicate is specified, but all of the following conditions are met:
– All columns in the table and in the query are in the index.
– At least one indexed column is not null.
• A query includes an ORDER BY on indexed non-nullable columns.
Branch Blocks
0..40
41..80
81..120
....
200..250
Leaf Blocks
8-29
Chapter 8
B-Tree Index Access Paths
The following statement queries the ID and name for departments in order of
department ID:
The following plan shows that the optimizer chose an index full scan:
---------------------------------------------------------------------------
-----
|Id | Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 2
(100)| |
| 1 | TABLE ACCESS BY INDEX ROWID| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
| 2 | INDEX FULL SCAN | DEPT_ID_PK | 27 | | 1 (0)|
00:00:01 |
---------------------------------------------------------------------------
-----
The database locates the first index leaf block, and then proceeds horizontally to the
right through the linked list of leaf nodes. For each index entry, the database obtains
the rowid from the entry, and then retrieves the table row specified by the rowid.
Because the index is sorted on department_id, the database avoids a separate
operation to sort the retrieved rows.
8-30
Chapter 8
B-Tree Index Access Paths
Note:
Unlike a full scan, a fast full scan cannot eliminate a sort operation because it
does not read the index in order.
See Also:
Oracle Database SQL Language Reference to learn more about the INDEX
hint
The following plan shows that the optimizer chose a fast full index scan:
8-31
Chapter 8
B-Tree Index Access Paths
See Also:
Oracle Database Concepts
8-32
Chapter 8
B-Tree Index Access Paths
logical index separately, and "skips" index blocks that do not meet the filter condition
on the non-leading column.
The customers table contains a column cust_gender whose values are either M or F.
While logged in to the database as user sh, you create a composite index on the
columns (cust_gender, cust_email) as follows:
Conceptually, a portion of the index might look as follows, with the gender value of F or
M as the leading edge of the index.
F,Wolf@company.example.com,rowid
F,Wolsey@company.example.com,rowid
F,Wood@company.example.com,rowid
F,Woodman@company.example.com,rowid
F,Yang@company.example.com,rowid
F,Zimmerman@company.example.com,rowid
M,Abbassi@company.example.com,rowid
M,Abbey@company.example.com,rowid
You run the following query for a customer in the sh.customers table:
SELECT *
FROM sh.customers
WHERE cust_email = 'Abbey@company.example.com';
The database can use a skip scan of the customers_gender_email index even though
cust_gender is not specified in the WHERE clause. In the sample index, the leading
column cust_gender has two possible values: F and M. The database logically splits
the index into two. One subindex has the key F, with entries in the following form:
F,Wolf@company.example.com,rowid
F,Wolsey@company.example.com,rowid
F,Wood@company.example.com,rowid
F,Woodman@company.example.com,rowid
F,Yang@company.example.com,rowid
F,Zimmerman@company.example.com,rowid
The second subindex has the key M, with entries in the following form:
M,Abbassi@company.example.com,rowid
M,Abbey@company.example.com,rowid
When searching for the record for the customer whose email is
Abbey@company.example.com, the database searches the subindex with the leading
8-33
Chapter 8
B-Tree Index Access Paths
value F first, and then searches the subindex with the leading value M. Conceptually,
the database processes the query as follows:
( SELECT *
FROM sh.customers
WHERE cust_gender = 'F'
AND cust_email = 'Abbey@company.example.com' )
UNION ALL
( SELECT *
FROM sh.customers
WHERE cust_gender = 'M'
AND cust_email = 'Abbey@company.example.com' )
-----------------------------------------------------------------------------------------
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)|Time|
-----------------------------------------------------------------------------------------
| 0|SELECT STATEMENT | | | |10(100)| |
| 1| TABLE ACCESS BY INDEX ROWID BATCHED| CUSTOMERS |33|6237| 10(0)|00:00:01|
|*2| INDEX SKIP SCAN | CUST_GENDER_EMAIL_IX |33| | 4(0)|00:00:01|
-----------------------------------------------------------------------------------------
2 - access("CUST_EMAIL"='Abbey@company.example.com')
filter("CUST_EMAIL"='Abbey@company.example.com')
See Also:
Oracle Database Concepts to learn more about skip scans
8-34
Chapter 8
B-Tree Index Access Paths
See Also:
Oracle Database SQL Language Reference
8-35
Chapter 8
B-Tree Index Access Paths
Banda,Amit,AAAVgdAALAAAABSABD
Bates,Elizabeth,AAAVgdAALAAAABSABI
Bell,Sarah,AAAVgdAALAAAABSABc
Bernstein,David,AAAVgdAALAAAABSAAz
Bissot,Laura,AAAVgdAALAAAABSAAd
Bloom,Harrison,AAAVgdAALAAAABSABF
Bull,Alexis,AAAVgdAALAAAABSABV
ABANDA,AAAVgdAALAAAABSABD
ABULL,AAAVgdAALAAAABSABV
ACABRIO,AAAVgdAALAAAABSABX
AERRAZUR,AAAVgdAALAAAABSAAv
AFRIPP,AAAVgdAALAAAABSAAV
AHUNOLD,AAAVgdAALAAAABSAAD
AHUTTON,AAAVgdAALAAAABSABL
8-36
Chapter 8
Bitmap Index Access Paths
8-37
Chapter 8
Bitmap Index Access Paths
101,AAAPvCAAFAAAAFaAAa
Nonunique B-tree Indexed data None In an entry of the index on the employees.last_name
combined with rowid column, the name and rowid combination
Smith,AAAPvCAAFAAAAFaAAa is the key, and there is no
data:
Smith,AAAPvCAAFAAAAFaAAa
M,low-rowid,high-rowid,1000101010101010
The database stores a bitmap index in a B-tree structure. The database can search
the B-tree quickly on the first part of the key, which is the set of attributes on which the
index is defined, and then obtain the corresponding rowid range and bitmap.
See Also:
• "Bitmap Storage"
• Oracle Database Concepts for an overview of bitmap indexes
• Oracle Database Data Warehousing Guide for more information about
bitmap indexes
8-38
Chapter 8
Bitmap Index Access Paths
In general, B-tree indexes are suitable for columns with high NDV and frequent DML
activity. For example, the optimizer might choose a B-tree index for a query of a
sales.amount column that returns few rows. In contrast, the customers.state and
customers.county columns are candidates for bitmap indexes because they have few
distinct values, are infrequently updated, and can benefit from efficient AND and OR
operations.
Bitmap indexes are a useful way to speed ad hoc queries in a data warehouse. They
are fundamental to star transformations. Specifically, bitmap indexes are useful in
queries that contain the following:
• Multiple conditions in the WHERE clause
Before the table itself is accessed, the database filters out rows that satisfy some,
but not all, conditions.
• AND, OR, and NOT operations on columns with low or medium NDV
Combining bitmap indexes makes these operations more efficient. The database
can merge bitmaps from bitmap indexes very quickly. For example, if bitmap
indexes exist on the customers.state and customers.county columns, then these
indexes can enormously improve the performance of the following query:
SELECT *
FROM customers
WHERE state = 'CA'
AND county = 'San Mateo'
The database can convert 1 values in the merged bitmap into rowids efficiently.
• The COUNT function
The database can scan the bitmap index without needing to scan the table.
• Predicates that select for null values
Unlike B-tree indexes, bitmap indexes can contain nulls. Queries that count the
number of nulls in a column can use the bitmap index without scanning the table.
• Columns that do not experience heavy DML
The reason is that one index key points to many rows. If a session modifies the
indexed data, then the database cannot lock a single bit in the bitmap: rather, the
database locks the entire index entry, which in practice locks the rows pointed to
by the bitmap. For example, if the county of residence for a specific customer
changes from San Mateo to Alameda, then the database must get exclusive access
to the San Mateo index entry and Alameda index entry in the bitmap. Rows
containing these two values cannot be modified until COMMIT.
See Also:
• "Star Transformation"
• Oracle Database SQL Language Reference to learn about the COUNT
function
8-39
Chapter 8
Bitmap Index Access Paths
Note:
The Hakan factor is an optimization used by the bitmap index algorithms to
limit the number of rows that Oracle Database assumes can be stored in a
single block. By artificially limiting the number of rows, the database reduces
the size of the bitmaps.
As shown in Table 8-4, bitmap indexes can include keys that consist entirely of null
values, unlike B-tree indexes. In Table 8-4, the null has a value of 1 for the 6th row in
the range, which means that the cust_marital_status value is null for the 6th row in
the range. Indexing nulls can be useful for some SQL statements, such as queries with
the aggregate function COUNT.
See Also:
Oracle Database Concepts to learn about rowid formats
8-40
Chapter 8
Bitmap Index Access Paths
The FROM and WHERE clause in the preceding CREATE statement represent the join
condition between the tables. The customers.cust_city column is the index key.
Each key value in the index represents a possible city in the customers table.
Conceptually, key values for the index might look as follows, with one bitmap
associated with each key value:
San Francisco 0 0 0 1 0 1 0 0 0 1 0 0 0 0 0 . . .
San Mateo 0 1 0 0 0 0 0 0 1 0 0 0 0 0 1 . . .
Smithville 1 0 0 0 1 0 0 1 0 0 1 0 1 0 0 . . .
.
.
.
Each bit in a bitmap corresponds to one row in the sales table. In the Smithville key,
the value 1 means that the first row in the sales table corresponds to a product sold to
a Smithville customer, whereas the value 0 means that the second row corresponds to
a product not sold to a Smithville customer.
Consider the following query of the number of separate sales to Smithville customers:
The following plan shows that the database reads the Smithville bitmap to derive the
number of Smithville sales (Step 4), thereby avoiding a join of the customers and
sales tables.
------------------------------------------------------------------------------------
8-41
Chapter 8
Bitmap Index Access Paths
4 - access("S"."SYS_NC00008$"='Smithville')
See Also:
Oracle Database Concepts to learn about the CREATE INDEX statement
8-42
Chapter 8
Bitmap Index Access Paths
The following plan shows that the database uses a range scan to find all key values
less than 1918 (Step 3), converts the 1 values in the bitmap to rowids (Step 2), and
then uses the rowids to obtain the rows from the customers table (Step 1):
-------------------------------------------------------------------------------------------
--
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
-------------------------------------------------------------------------------------------
--
| 0| SELECT STATEMENT | | | |421
(100)| |
| 1| TABLE ACCESS BY INDEX ROWID BATCHED| CUSTOMERS |3604|68476|421 (1)|
00:00:01 |
| 2| BITMAP CONVERSION TO ROWIDS | | | |
| |
|*3| BITMAP INDEX RANGE SCAN | CUSTOMERS_YOB_BIX| | |
| |
-------------------------------------------------------------------------------------------
8-43
Chapter 8
Bitmap Index Access Paths
--
The database only needs to process a single bitmap. For example, the following table
represents the bitmap index (in two bitmap pieces) for the value widowed in the
sh.customers.cust_marital_status column. To satisfy a query of customers with the
status widowed, the database can search for each 1 value in the widowed bitmap and
find the rowid of the corresponding row.
8-44
Chapter 8
Bitmap Index Access Paths
SELECT *
FROM customers
WHERE cust_marital_status = 'Widowed';
The following plan shows that the database reads the entry with the Widowed key in the
customers bitmap index (Step 3), converts the 1 values in the bitmap to rowids (Step
2), and then uses the rowids to obtain the rows from the customers table (Step 1):
3 - access("CUST_MARITAL_STATUS"='Widowed')
8-45
Chapter 8
Bitmap Index Access Paths
See Also:
"Index Range Scans"
See Also:
"Index Range Scans"
8-46
Chapter 8
Bitmap Index Access Paths
The following plan shows that the database obtains all bitmaps for
cust_year_of_birth keys lower than 1918 (Step 3), converts the bitmaps to rowids
(Step 2), and then fetches the rows (Step 1):
8-47
Chapter 8
Bitmap Index Access Paths
1917 1 0 1 0 0 0 0 0 0 0 0 0 0 1
1916 0 1 0 0 0 0 0 0 0 0 0 0 0 0
1915 0 0 0 0 0 0 0 0 1 0 0 0 0 0
------------------------------------
merged: 1 1 1 0 0 0 0 0 1 0 0 0 0 1
The 1 values in resulting bitmap correspond to rows that contain the values 1915, 1916,
or 1917.
The following plan shows that the database obtains all bitmaps for
cust_year_of_birth keys lower than 1918 (Step 6), and then merges these bitmaps
using OR logic to create a single bitmap (Step 5). The database obtains a single bitmap
for the cust_gender key of F (Step 4), and then performs an AND operation on these
two bitmaps. The result is a single bitmap that contains 1 values for the requested
rows (Step 3).
8-48
Chapter 8
Table Cluster Access Paths
(100)| |
| 1| TABLE ACCESS BY INDEX ROWID BATCHED|CUSTOMERS |1802|37842|288 (1)|
00:00:01|
| 2| BITMAP CONVERSION TO ROWIDS | | | |
| |
| 3| BITMAP AND | | | |
| |
|*4| BITMAP INDEX SINGLE VALUE |CUSTOMERS_GENDER_BIX| | |
| |
| 5| BITMAP MERGE | | | |
| |
|*6| BITMAP INDEX RANGE SCAN |CUSTOMERS_YOB_BIX | | |
| |
-------------------------------------------------------------------------------------------
--
• Cluster Scans
An index cluster is a table cluster that uses an index to locate data.
• Hash Scans
A hash cluster is like an indexed cluster, except the index key is replaced with a
hash function. No separate cluster index exists.
See Also:
Oracle Database Concepts for an overview of table clusters
8-49
Chapter 8
Table Cluster Access Paths
30,AADAAAA9d
When a user requests rows in the cluster, the database scans the index to obtain the
DBAs of the blocks containing the rows. Oracle Database then locates the rows based
on these DBAs.
8-50
Chapter 8
Table Cluster Access Paths
SELECT *
FROM employees2
WHERE department_id = 30;
To perform the scan, Oracle Database first obtains the rowid of the row describing
department 30 by scanning the cluster index (Step 2). Oracle Database then locates
the rows in employees2 using this rowid (Step 1).
---------------------------------------------------------------------------
-----
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | | | | 2
(100)| |
| 1| TABLE ACCESS CLUSTER| EMPLOYEES2 | 6 | 798 | 2 (0)|
00:00:01|
|*2| INDEX UNIQUE SCAN |IDX_EMP_DEPT_CLUSTER| 1 | | 1 (0)|
00:00:01|
---------------------------------------------------------------------------
-----
2 - access("DEPARTMENT_ID"=30)
See Also:
Oracle Database Concepts to learn about indexed clusters
8-51
Chapter 8
Table Cluster Access Paths
In a hash cluster, the data is the index. The database uses a hash scan to locate rows
in a hash cluster based on a hash value.
This section contains the following topics:
• When the Optimizer Considers a Hash Scan
The database considers a hash scan when a query accesses a table in a hash
cluster.
• How a Hash Scan Works
In a hash cluster, all rows with the same hash value are stored in the same data
block.
• Hash Scans: Example
This example hashes the employees and departments tables on the
department_id column, and then queries the cluster for a single department.
SELECT *
FROM employees2
WHERE department_id = 30
8-52
Chapter 8
Table Cluster Access Paths
To perform a hash scan, Oracle Database first obtains the hash value by applying a
hash function to the key value 30, and then uses this hash value to scan the data
blocks and retrieve the rows (Step 1).
----------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost |
----------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 1 |
|* 1 | TABLE ACCESS HASH| EMPLOYEES2 | 10 | 1330 | |
----------------------------------------------------------------
1 - access("DEPARTMENT_ID"=30)
See Also:
Oracle Database Concepts to learn about hash clusters
8-53
9
Joins
Oracle Database provides several optimizations for joining row sets.
This chapter contains the following topics:
• About Joins
A join combines the output from exactly two row sources, such as tables or views,
and returns one row source. The returned row source is the data set.
• Join Methods
A join method is the mechanism for joining two row sources.
• Join Types
A join type is determined by the type of join condition.
• Join Optimizations
Join optimizations enable joins to be more efficient.
• Join Trees
Typically, a join tree is represented as an upside-down tree structure.
• How the Optimizer Executes Join Statements
The database joins pairs of row sources. When multiple tables exist in the FROM
clause, the optimizer must determine which join operation is most efficient for each
pair.
• How the Optimizer Chooses Execution Plans for Joins
When determining the join order and method, the optimizer goal is to reduce the
number of rows early so it performs less work throughout the execution of the SQL
statement.
9-1
Chapter 9
About Joins
See Also:
• "Cartesian Joins"
• Oracle Database SQL Language Reference for a concise discussion of
joins in Oracle SQL
result set
table1 table2
The input of a join can be the result set from a previous join. If the right child of every
internal node of a join tree is a table, then the tree is a left deep join tree, as shown in
the following example. Most join trees are left deep joins.
result set
table4
table3
table1 table2
9-2
Chapter 9
About Joins
If the left child of every internal node of a join tree is a table, then the tree is called a
right deep join tree, as shown in the following diagram.
result set
table1
table2
table3 table4
If the left or the right child of an internal node of a join tree can be a join node, then the
tree is called a bushy join tree. In the following example, table4 is a right child of a join
node, table1 is the left child of a join node, and table2 is the left child of a join node.
result set
table4
table1
table2 table3
In yet another variation, both inputs of a join are the results of a previous join.
9-3
Chapter 9
About Joins
9-4
Chapter 9
Join Methods
tables from the subquery must come after those tables in the outer query block to
which they were connected or correlated. However, hash antijoins and semijoins
are able to override this ordering condition in certain circumstances.
The optimizer estimates the cost of a query plan by computing the estimated I/Os and
CPU. These I/Os have specific costs associated with them: one cost for a single block
I/O, and another cost for multiblock I/Os. Also, different functions and expressions
have CPU costs associated with them. The optimizer determines the total cost of a
query plan using these metrics. These metrics may be influenced by many initialization
parameter and session settings at compile time, such as the
DB_FILE_MULTI_BLOCK_READ_COUNT setting, system statistics, and so on.
Table 9-2 Sample Costs for Join of date_dim and lineorder Tables
See Also:
9-5
Chapter 9
Join Methods
Depending on the statistics, the optimizer chooses the method with the lowest
estimated cost. As shown in Figure 9-5, each join method has two children: the driving
(also called outer) row source and the driven-to (also called inner) row source.
Join Method
(Nested Loops, Hash
Join, or Sort Merge)
9-6
Chapter 9
Join Methods
Note:
The number of rows expected from the join is what drives the optimizer
decision, not the size of the underlying tables. For example, a query might
join two tables of a billion rows each, but because of the filters the optimizer
expects data sets of 5 rows each.
In general, nested loops joins work best on small tables with indexes on the join
conditions. If a row source has only one row, as with an equality lookup on a primary
key value (for example, WHERE employee_id=101), then the join is a simple lookup. The
optimizer always tries to put the smallest row source first, making it the driving table.
Various factors enter into the optimizer decision to use nested loops. For example, the
database may read several rows from the outer row source in a batch. Based on the
number of rows retrieved, the optimizer may choose either a nested loop or a hash join
to the inner row source. For example, if a query joins departments to driving table
employees, and if the predicate specifies a value in employees.last_name, then the
database might read enough entries in the index on last_name to determine whether
an internal threshold is passed. If the threshold is not passed, then the optimizer picks
a nested loop join to departments, and if the threshold is passed, then the database
performs a hash join, which means reading the rest of employees, hashing it into
memory, and then joining to departments.
If the access path for the inner loop is not dependent on the outer loop, then the result
can be a Cartesian product: for every iteration of the outer loop, the inner loop
produces the same set of rows. To avoid this problem, use other join methods to join
two independent row sources.
See Also:
• "Table 19-2"
• "Adaptive Query Plans"
9-7
Chapter 9
Join Methods
For example, if a query joins employees and departments, then a nested loop in
pseudocode might be:
The inner loop is executed for every row of the outer loop. The employees table is the
"outer" data set because it is in the exterior for loop. The outer table is sometimes
called a driving table. The departments table is the "inner" data set because it is in the
interior for loop.
NESTED LOOPS
outer_loop
inner_loop
3. For every fetch request from the client, the basic process is as follows:
a. Fetch a row from the outer row source
b. Probe the inner row source to find rows that match the predicate criteria
c. Repeat the preceding steps until all rows are obtained by the fetch request
Sometimes the database sorts rowids to obtain a more efficient buffer access
pattern.
SELECT STATEMENT
NESTED LOOPS 3
9-8
Chapter 9
Join Methods
NESTED LOOPS 1
OUTER LOOP 1.1
INNER LOOP 1.2
NESTED LOOPS 2
OUTER LOOP 2.1 - Row source generated by NESTED LOOPS 1
INNER LOOP 2.2
NESTED LOOPS 3
OUTER LOOP 3.1 - Row source generated by NESTED LOOPS 2
INNER LOOP 3.2
The plan reveals that the optimizer chose two nested loops (Step 1 and Step 2) to
access the data:
9-9
Chapter 9
Join Methods
----------------------------------------------------------------------------------
|Id| Operation |Name |Rows|Bytes|Cost(%CPU)|Time|
----------------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | |5 (100)| |
| 1| NESTED LOOPS | | | | | |
| 2| NESTED LOOPS | | 3|102|5 (0)|00:00:01|
| 3| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES | 3| 54|2 (0)|00:00:01|
|*4| INDEX RANGE SCAN | EMP_NAME_IX | 3| |1 (0)|00:00:01|
|*5| INDEX UNIQUE SCAN | DEPT_ID_PK | 1| |0 (0)| |
| 6| TABLE ACCESS BY INDEX ROWID | DEPARTMENTS | 1| 16|1 (0)|00:00:01|
----------------------------------------------------------------------------------
Abel,employees_rowid
Ande,employees_rowid
Atkinson,employees_rowid
Austin,employees_rowid
b. Using the rowids from the previous step, the database retrieves a batch of
rows from the employees table (Step 3). For example:
Abel,Ellen,80
Abel,John,50
These rows become the outer row source for the innermost nested loop.
The batch step is typically part of adaptive execution plans. To determine
whether a nested loop is better than a hash join, the optimizer needs to
determine many rows come back from the row source. If too many rows are
returned, then the optimizer switches to a different join method.
c. For each row in the outer row source, the database scans the dept_id_pk
index to obtain the rowid in departments of the matching department ID (Step
5), and joins it to the employees rows. For example:
Abel,Ellen,80,departments_rowid
Ande,Sundar,80,departments_rowid
Atkinson,Mozhe,50,departments_rowid
Austin,David,60,departments_rowid
9-10
Chapter 9
Join Methods
These rows become the outer row source for the outer nested loop (Step 1).
2. The database iterates through the outer nested loop as follows:
a. The database reads the first row in outer row source.
For example:
Abel,Ellen,80,departments_rowid
b. The database uses the departments rowid to retrieve the corresponding row
from departments (Step 6), and then joins the result to obtain the requested
values (Step 1).
For example:
Abel,Ellen,80,Sales
c. The database reads the next row in the outer row source, uses the
departments rowid to retrieve the corresponding row from departments (Step
6), and iterates through the loop until all rows are retrieved.
The result set has the following form:
Abel,Ellen,80,Sales
Ande,Sundar,80,Sales
Atkinson,Mozhe,50,Shipping
Austin,David,60,IT
-------------------------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes|Cost%CPU| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 19 | 722 | 3 (0)|00:00:01|
| 1 | NESTED LOOPS | | | | | |
| 2 | NESTED LOOPS | | 19 | 722 | 3 (0)|00:00:01|
|* 3 | TABLE ACCESS FULL | DEPARTMENTS | 2 | 32 | 2 (0)|00:00:01|
|* 4 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 10 | | 0 (0)|00:00:01|
9-11
Chapter 9
Join Methods
In this case, rows from the hr.departments table form the outer row source (Step 3) of
the inner nested loop (Step 2). The index emp_department_ix is the inner row source
(Step 4) of the inner nested loop. The results of the inner nested loop form the outer
row source (Row 2) of the outer nested loop (Row 1). The hr.employees table is the
outer row source (Row 5) of the outer nested loop.
For each fetch request, the basic process is as follows:
1. The database iterates through the inner nested loop (Step 2) to obtain the rows
requested in the fetch:
a. The database reads the first row of departments to obtain the department IDs
for departments named Marketing or Sales (Step 3). For example:
Marketing,20
This row set is the outer loop. The database caches the data in the PGA.
b. The database scans emp_department_ix, which is an index on the employees
table, to find employees rowids that correspond to this department ID (Step 4),
and then joins the result (Step 2).
The result set has the following form:
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
.
.
.
Sales,80,employees_rowid
Sales,80,employees_rowid
Sales,80,employees_rowid
.
9-12
Chapter 9
Join Methods
.
.
These rows become the outer row source for the outer nested loop (Step 1).
This row set is cached in the PGA.
2. The database organizes the rowids obtained in the previous step so that it can
more efficiently access them in the cache.
3. The database begins iterating through the outer nested loop as follows:
a. The database retrieves the first row from the row set obtained in the previous
step, as in the following example:
Marketing,20,employees_rowid
b. Using the rowid, the database retrieves a row from employees to obtain the
requested values (Step 1), as in the following example:
Michael,Hartstein,13000,Marketing
c. The database retrieves the next row from the row set, uses the rowid to probe
employees for the matching row, and iterates through the loop until all rows are
retrieved.
The result set has the following form:
Michael,Hartstein,13000,Marketing
Pat,Fay,6000,Marketing
John,Russell,14000,Sales
Karen,Partners,13500,Sales
Alberto,Errazuriz,12000,Sales
.
.
.
In some cases, a second join row source is not allocated, and the execution plan looks
the same as it did before Oracle Database 11g. The following list describes such
cases:
• All of the columns needed from the inner side of the join are present in the index,
and there is no table access required. In this case, Oracle Database allocates only
one join row source.
• The order of the rows returned might be different from the order returned in
releases earlier than Oracle Database 12c. Thus, when Oracle Database tries to
preserve a specific ordering of the rows, for example to eliminate the need for an
ORDER BY sort, Oracle Database might use the original implementation for nested
loops joins.
• The OPTIMIZER_FEATURES_ENABLE initialization parameter is set to a release before
Oracle Database 11g. In this case, Oracle Database uses the original
implementation for nested loops joins.
9-13
Chapter 9
Join Methods
In releases before Oracle Database 11g, the execution plan for this query might
appear as follows:
-------------------------------------------------------------------------------------------
-----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
-------------------------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 19 | 722 | 3 (0)|
00:00:01 |
| 1 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES | 10 | 220 | 1 (0)|
00:00:01 |
| 2 | NESTED LOOPS | | 19 | 722 | 3 (0)|
00:00:01 |
|* 3 | TABLE ACCESS FULL | DEPARTMENTS | 2 | 32 | 2 (0)|
00:00:01 |
|* 4 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 10 | | 0 (0)|
00:00:01 |
-------------------------------------------------------------------------------------------
-----
This row set is the outer loop. The database caches the row in the PGA.
b. The database scans emp_department_ix, which is an index on the
employees.department_id column, to find employees rowids that correspond
to this department ID (Step 4), and then joins the result (Step 2).
9-14
Chapter 9
Join Methods
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
Marketing,20,employees_rowid
.
.
.
Sales,80,employees_rowid
Sales,80,employees_rowid
Sales,80,employees_rowid
.
.
.
2. Depending on the circumstances, the database may organize the cached rowids
obtained in the previous step so that it can more efficiently access them.
3. For each employees rowid in the result set generated by the nested loop, the
database retrieves a row from employees to obtain the requested values (Step 1).
Thus, the basic process is to read a rowid and retrieve the matching employees
row, read the next rowid and retrieve the matching employees row, and so on.
Conceptually, the result set has the following form:
Michael,Hartstein,13000,Marketing
Pat,Fay,6000,Marketing
John,Russell,14000,Sales
Karen,Partners,13500,Sales
Alberto,Errazuriz,12000,Sales
.
.
.
9-15
Chapter 9
Join Methods
table as the inner table. The index is optional. If no index is specified, then the nested
loops join uses an index with at least one join predicate as the index key.
Example 9-3 Nested Loops Hint
Assume that the optimizer chooses a hash join for the following query:
---------------------------------------------------------------------------
---
|Id | Operation | Name | Rows | Bytes |Cost(%CPU)|
Time |
---------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | | | | 5
(100)| |
|*1 | HASH JOIN | | 106 | 2862 | 5 (20)|
00:00:01 |
| 2 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
| 3 | TABLE ACCESS FULL| EMPLOYEES | 107 | 1177 | 2 (0)|
00:00:01 |
---------------------------------------------------------------------------
---
To force a nested loops join using departments as the inner table, add the USE_NL hint
as in the following query:
---------------------------------------------------------------------------
-----
| Id | Operation | Name | Rows | Bytes |Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 34
(100)| |
| 1 | NESTED LOOPS | | 106 | 2862 | 34 (3)|
00:00:01 |
| 2 | TABLE ACCESS FULL| EMPLOYEES | 107 | 1177 | 2 (0)|
00:00:01 |
|* 3 | TABLE ACCESS FULL| DEPARTMENTS | 1 | 16 | 0
(0)| |
9-16
Chapter 9
Join Methods
---------------------------------------------------------------------------
-----
3 - filter("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
De Haan,90
2. For the row obtained in the previous step, the database scans departments to find
the department name that matches the employees department ID (Step 3), and
joins the result (Step 1). For example:
De Haan,Executive
3. The database retrieves the next row in employees, retrieves the matching row from
departments, and then repeats this process until all rows are retrieved.
The result set has the following form:
De Haan,Executive
Kochnar,Executive
Baer,Public Relations
King,Executive
.
.
.
See Also:
• "Guidelines for Join Order Hints" to learn more about the USE_NL hint
• Oracle Database SQL Language Reference to learn about the USE_NL
hint
9-17
Chapter 9
Join Methods
9-18
Chapter 9
Join Methods
The database applies the hash function to each department_id in the table,
generating a hash value for each. For this illustration, the hash table has 5 slots (it
could have more or less). Because n is 5, the possible hash values range from 1 to 5.
The hash functions might generate the following values for the department IDs:
f(10) = 4
f(20) = 1
f(30) = 4
f(40) = 2
f(50) = 5
Note that the hash function happens to generate the same hash value of 4 for
departments 10 and 30. This is known as a hash collision. In this case, the database
puts the records for departments 10 and 30 in the same slot, using a linked list.
Conceptually, the hash table looks as follows:
1 20,Marketing,201,1800
2 40,Human Resources,203,2400
3
4 10,Administration,200,1700 -> 30,Purchasing,114,1700
5 50,Shipping,121,1500
9-19
Chapter 9
Join Methods
2. The database probes the second data set, called the probe table, using
whichever access mechanism has the lowest cost.
Typically, the database performs a full scan of both the smaller and larger data
set. The algorithm in pseudocode might look as follows:
For each row retrieved from the larger data set, the database does the following:
a. Applies the same hash function to the join column or columns to calculate the
number of the relevant slot in the hash table.
For example, to probe the hash table for department ID 30, the database
applies the hash function to 30, which generates the hash value 4.
b. Probes the hash table to determine whether rows exists in the slot.
If no rows exist, then the database processes the next row in the larger data
set. If rows exist, then the database proceeds to the next step.
c. Checks the join column or columns for a match. If a match occurs, then the
database either reports the rows or passes them to the next step in the plan,
and then processes the next row in the larger data set.
If multiple rows exist in the hash table slot, the database walks through the
linked list of rows, checking each one. For example, if department 30 hashes
to slot 4, then the database checks each row until it finds 30.
Example 9-4 Hash Joins
An application queries the oe.orders and oe.order_items tables, joining on the
order_id column.
9-20
Chapter 9
Join Methods
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 665 | 13300 | 8 (25)|
|* 1 | HASH JOIN | | 665 | 13300 | 8 (25)|
| 2 | TABLE ACCESS FULL | ORDERS | 105 | 840 | 4 (25)|
| 3 | TABLE ACCESS FULL | ORDER_ITEMS | 665 | 7980 | 4 (25)|
--------------------------------------------------------------------------
Because the orders table is small relative to the order_items table, which is 6 times
larger, the database hashes orders. In a hash join, the data set for the build table
always appears first in the list of operations (Step 2). In Step 3, the database performs
a full scan of the larger order_items later, probing the hash table for each row.
9.2.2.3 How Hash Joins Work When the Hash Table Does Not Fit in the PGA
The database must use a different technique when the hash table does not fit entirely
in the PGA. In this case, the database uses a temporary space to hold portions (called
partitions) of the hash table, and sometimes portions of the larger table that probes the
hash table.
The basic process is as follows:
1. The database performs a full scan of the smaller data set, and then builds an array
of hash buckets in both the PGA and on disk.
When the PGA hash area fills up, the database finds the largest partition within the
hash table and writes it to temporary space on disk. The database stores any new
row that belongs to this on-disk partition on disk, and all other rows in the PGA.
Thus, part of the hash table is in memory and part of it on disk.
2. The database takes a first pass at reading the other data set.
For each row, the database does the following:
a. Applies the same hash function to the join column or columns to calculate the
number of the relevant hash bucket.
b. Probes the hash table to determine whether rows exist in the bucket in
memory.
If the hashed value points to a row in memory, then the database completes
the join and returns the row. If the value points to a hash partition on disk,
however, then the database stores this row in the temporary tablespace, using
the same partitioning scheme used for the original data set.
3. The database reads each on-disk temporary partition one by one
4. The database joins each partition row to the row in the corresponding on-disk
temporary partition.
9-21
Chapter 9
Join Methods
See Also:
MERGE JOIN
9-22
Chapter 9
Join Methods
9-23
Chapter 9
Join Methods
For example, the following table shows sorted values in two data sets: temp_ds1 and
temp_ds2.
temp_ds1 temp_ds2
10 20
20 20
30 40
40 40
50 40
60 40
70 40
. 60
. 70
. 70
As shown in the following table, the database begins by reading 10 in temp_ds1, and
then reads the first value in temp_ds2. Because 20 in temp_ds2 is higher than 10 in
temp_ds1, the database stops reading temp_ds2.
The database proceeds to the next value in temp_ds1, which is 20. The database
proceeds through temp_ds2 as shown in the following table.
9-24
Chapter 9
Join Methods
The database proceeds to the next row in temp_ds1, which is 30. The database starts
at the number of its last match, which was 20, and then proceeds through temp_ds2
looking for a match, as shown in the following table.
The database proceeds to the next row in temp_ds1, which is 40. As shown in the
following table, the database starts at the number of its last match in temp_ds2, which
was 20, and then proceeds through temp_ds2 looking for a match.
9-25
Chapter 9
Join Methods
The database continues in this way until it has matched the final 70 in temp_ds2. This
scenario demonstrates that the database, as it reads through temp_ds1, does not need
to read every row in temp_ds2. This is an advantage over a nested loops join.
A query of DBMS_XPLAN.DISPLAY_CURSOR shows that the plan uses a sort merge join:
---------------------------------------------------------------------------
-----
|Id| Operation | Name |Rows|Bytes |Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | | | |
5(100)| |
| 1| MERGE JOIN | |106 | 4028 | 5 (20)|
00:00:01 |
| 2| TABLE ACCESS BY INDEX ROWID| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
| 3| INDEX FULL SCAN | DEPT_ID_PK | 27 | | 1 (0)|
00:00:01 |
|*4| SORT JOIN | |107 | 2354 | 3 (34)|
00:00:01 |
| 5| TABLE ACCESS FULL | EMPLOYEES |107 | 2354 | 2 (0)|
00:00:01 |
---------------------------------------------------------------------------
-----
4 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
filter("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
9-26
Chapter 9
Join Methods
The two data sets are the departments table and the employees table. Because an
index orders the departments table by department_id, the database can read this
index and avoid a sort (Step 3). The database only needs to sort the employees table
(Step 4), which is the most CPU-intensive operation.
Example 9-6 Sort Merge Join Without an Index
You join the employees and departments tables on the department_id column,
ordering the rows on department_id as follows. In this example, you specify NO_INDEX
and USE_MERGE to force the optimizer to choose a sort merge:
A query of DBMS_XPLAN.DISPLAY_CURSOR shows that the plan uses a sort merge join:
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 6
(100)| |
| 1 | MERGE JOIN | | 106 | 9540 | 6 (34)|
00:00:01|
| 2 | SORT JOIN | | 27 | 567 | 3 (34)|
00:00:01|
| 3 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 567 | 2 (0)|
00:00:01|
|*4 | SORT JOIN | | 107 | 7383 | 3 (34)|
00:00:01|
| 5 | TABLE ACCESS FULL| EMPLOYEES | 107 | 7383 | 2 (0)|
00:00:01|
---------------------------------------------------------------------------
-----
4 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
filter("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
9-27
Chapter 9
Join Types
In some situations it may make sense to override the optimizer with the USE_MERGE
hint. For example, the optimizer can choose a full scan on a table and avoid a sort
operation in a query. However, there is an increased cost because a large table is
accessed through an index and single block reads, as opposed to faster access
through a full table scan.
See Also:
Oracle Database SQL Language Reference to learn about the USE_MERGE
hint
9-28
Chapter 9
Join Types
• Band Joins
A band join is a special type of nonequijoin in which key values in one data set
must fall within the specified range (“band”) of the second data set. The same
table can serve as both the first and second data sets.
9.3.1.1 Equijoins
An equijoin is an inner join whose join condition contains an equality operator.
The following example is an equijoin because the join condition contains only an
equality operator:
9.3.1.2 Nonequijoins
A nonequijoin is an inner join whose join condition contains an operator that is not an
equality operator.
The following query lists all employees whose hire date occurred when employee 176
(who is listed in job_history because he changed jobs in 2007) was working at the
company:
In the preceding example, the condition joining employees and job_history does not
contain an equality operator, so it is a nonequijoin. Nonequijoins are relatively rare.
Note that a hash join requires at least a partial equijoin. The following SQL script
contains an equality join condition (e1.empno = e2.empno) and a nonequality
condition:
The optimizer chooses a hash join for the preceding query, as shown in the following
plan:
Execution Plan
----------------------------------------------------------
9-29
Chapter 9
Join Types
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 174 | 5 (20)| 00:00:01 |
|* 1 | HASH JOIN | | 1 | 174 | 5 (20)| 00:00:01 |
| 2 | TABLE ACCESS FULL| EMP | 14 | 1218 | 2 (0)| 00:00:01 |
| 3 | TABLE ACCESS FULL| EMP | 14 | 1218 | 2 (0)| 00:00:01 |
---------------------------------------------------------------------------
1 - access("E1"."EMPNO"="E2"."EMPNO")
filter("E1"."HIREDATE">=INTERNAL_FUNCTION("E2"."HIREDATE")-1 AND
"E1"."HIREDATE"<=INTERNAL_FUNCTION("E2"."HIREDATE")+1)
SELECT e1.last_name ||
' has salary between 100 less and 100 more than ' ||
e2.last_name AS "SALARY COMPARISON"
FROM employees e1,
employees e2
WHERE e1.salary
BETWEEN e2.salary - 100
AND e2.salary + 100;
SALARY COMPARISON
-------------------------------------------------------------
King has salary between 100 less and 100 more than King
Kochhar has salary between 100 less and 100 more than Kochhar
Kochhar has salary between 100 less and 100 more than De Haan
De Haan has salary between 100 less and 100 more than Kochhar
De Haan has salary between 100 less and 100 more than De Haan
Russell has salary between 100 less and 100 more than Russell
9-30
Chapter 9
Join Types
Partners has salary between 100 less and 100 more than Partners
...
------------------------------------------
PLAN_TABLE_OUTPUT
------------------------------------------
------------------------------------------
| Id | Operation | Name |
------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | MERGE JOIN | |
| 2 | SORT JOIN | |
| 3 | TABLE ACCESS FULL | EMPLOYEES |
|* 4 | FILTER | |
|* 5 | SORT JOIN | |
| 6 | TABLE ACCESS FULL| EMPLOYEES |
------------------------------------------
In this plan, Step 2 sorts the e1 row source, and Step 5 sorts the e2 row source. The
sorted row sources are illustrated in the following table.
The join begins by iterating through the sorted input (e1), which is the left branch of the
join, corresponding to Step 2 of the plan. The original query contains two predicates:
• e1.sal >= e2.sal–100, which is the Step 5 filter
• e1.sal >= e2.sal+100, which is the Step 4 filter
For each iteration of the sorted row source e1, the database iterates through row
source e2, checking every row against Step 5 filter e1.sal >= e2.sal–100. If the row
passes the Step 5 filter, then the database sends it to the Step 4 filter, and then
proceeds to test the next row in e2 against the Step 5 filter. However, if a row fails the
Step 5 filter, then the scan of e2 stops, and the database proceeds through the next
iteration of e1.
9-31
Chapter 9
Join Types
The following table shows the first iteration of e1, which begins with 24000 (King) in
data set e1. The database determines that the first row in e2, which is 24000 (King),
passes the Step 5 filter. The database then sends the row to the Step 4 filter, e1.sal
<= w2.sal+100, which also passes. The database sends this row to the MERGE row
source. Next, the database checks 17000 (Kochhar) against the Step 5 filter, which
also passes. However, the row fails the Step 4 filter, and is discarded. The database
proceeds to test 17000 (De Haan) against the Step 5 filter.
Table 9-9 First Iteration of e1: Separate SORT JOIN and FILTER
Scan e2 Step 5 Filter (e1.sal >= e2.sal–100) Step 4 Filter (e1.sal <= e2.sal+100)
24000 (King) Pass because 24000 >= 23900. Pass because 24000 <= 24100.
Send to Step 4 filter. Return row for merging.
17000 Pass because 24000 >= 16900. Fail because 24000 <=17100 is false.
(Kochhar) Send to Step 4 filter. Discard row. Scan next row in e2.
17000 (De Pass because 24000 >= 16900. Fail because 24000 <=17100 is false.
Haan) Send to Step 4 filter. Discard row. Scan next row in e2.
14000 (Russell) Pass because 24000 >= 13900. Fail because 24000 <=14100 is false.
Send to Step 4 filter. Discard row. Scan next row in e2.
13500 Pass because 24000 >= 13400. Fail because 24000 <=13600 is false.
(Partners) Send to Step 4 filter. Discard row. Scan next row in e2.
As shown in the preceding table, every e2 row necessarily passes the Step 5 filter
because the e2 salaries are sorted in descending order. Thus, the Step 5 filter always
sends the row to the Step 4 filter. Because the e2 salaries are sorted in descending
order, the Step 4 filter necessarily fails every row starting with 17000 (Kochhar). The
inefficiency occurs because the database tests every subsequent row in e2 against the
Step 5 filter, which necessarily passes, and then against the Step 4 filter, which
necessarily fails.
Example 9-8 Query With Band Join Optimization
Starting in Oracle Database 12c Release 2 (12.2), the database optimizes the band
join by using the following plan, which does not have a separate FILTER operation:
------------------------------------------
PLAN_TABLE_OUTPUT
------------------------------------------
| Id | Operation | Name |
------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | MERGE JOIN | |
| 2 | SORT JOIN | |
| 3 | TABLE ACCESS FULL | EMPLOYEES |
|* 4 | SORT JOIN | |
| 5 | TABLE ACCESS FULL | EMPLOYEES |
------------------------------------------
9-32
Chapter 9
Join Types
filter(("E1"."SALARY"<="E2"."SALARY"+100 AND
INTERNAL_FUNCTION("E1"."SALARY")>="E2"."SALARY"-100))
The difference is that Step 4 uses Boolean AND logic for the two predicates to create a
single filter. Instead of checking a row against one filter, and then sending it to a
different row source for checking against a second filter, the database performs one
check against one filter. If the check fails, then processing stops.
In this example, the query begins the first iteration of e1, which begins with 24000
(King). The following figure represents the range. e2 values below 23900 and above
24100 fall outside the range.
The following table shows that the database tests the first row of e2, which is 24000
(King), against the Step 4 filter. The row passes the test, so the database sends the
row to be merged. The next row in e2 is 17000 (Kochhar). This row falls outside of the
range (band) and thus does not satisfy the filter predicate, so the database stops
testing e2 rows in this iteration. The database stops testing because the descending
sort of e2 ensures that all subsequent rows in e2 fail the filter test. Thus, the database
can proceed to the second iteration of e1.
Scan e2 Filter 4 (e1.sal >= e2.sal – 100) AND (e1.sal <= e2.sal + 100)
24000 (King) Passes test because it is true that (24000 >= 23900) AND (24000
<= 24100).
Send row to MERGE. Test next row.
17000 (Kochhar) Fails test because it is false that (24000 >= 16900) AND (24000 <=
17100).
Stop scanning e2. Begin next iteration of e1.
In this way, the band join optimization eliminates unnecessary processing. Instead of
scanning every row in e2 as in the unoptimized case, the database scans only the
minimum two rows.
9-33
Chapter 9
Join Types
Outer joins require the outer-joined table to be the driving table. In the preceding
example, employees is the driving table, and departments is the driven-to table.
9-34
Chapter 9
Join Types
---------------------------------------------------------------------------
-----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 7
(100)| |
9-35
Chapter 9
Join Types
2 - access("C"."CUSTOMER_ID"="O"."CUSTOMER_ID")
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
----
3 - filter("C"."CREDIT_LIMIT">1000)
4 - filter("O"."CUSTOMER_ID">0)
The query looks for customers which satisfy various conditions. An outer join returns
NULL for the inner table columns along with the outer (preserved) table rows when it
does not find any corresponding rows in the inner table. This operation finds all the
customers rows that do not have any orders rows.
customers.customer_id = orders.customer_id(+)
---------------------------------------------------------------------------
-
| Id | Operation | Name | Rows | Bytes | Cost
(%CPU)|
---------------------------------------------------------------------------
-
| 0 | SELECT STATEMENT | | 144 | 4608 | 16
(32)|
| 1 | HASH GROUP BY | | 144 | 4608 | 16
(32)|
9-36
Chapter 9
Join Types
9-37
Chapter 9
Join Types
joins join tables together, yet show rows with no corresponding rows in the joined
tables.
Example 9-11 Full Outer Join
The following query retrieves all departments and all employees in each department,
but also includes:
• Any employees without departments
• Any departments without employees
DEPARTMENT_ID EMPLOYEE_ID
------------- -----------
10 200
20 201
20 202
30 114
30 115
30 116
...
270
280
178
207
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows |Bytes |Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 122 | 4758 | 6 (34)| 00:0
0:01 |
| 1 | SORT ORDER BY | | 122 | 4758 | 6 (34)| 00:0
0:01 |
| 2 | VIEW | VW_FOJ_0 | 122 | 4758 | 5 (20)| 00:0
0:01 |
9-38
Chapter 9
Join Types
HASH JOIN FULL OUTER is included in the preceding plan (Step 3), indicating that the
query uses the hash full outer join execution method. Typically, when the full outer join
condition between two tables is an equijoin, the hash full outer join execution method
is possible, and Oracle Database uses it automatically.
To instruct the optimizer to consider using the hash full outer join execution method,
apply the NATIVE_FULL_OUTER_JOIN hint. To instruct the optimizer not to consider using
the hash full outer join execution method, apply the NO_NATIVE_FULL_OUTER_JOIN hint.
The NO_NATIVE_FULL_OUTER_JOIN hint instructs the optimizer to exclude the native
execution method when joining each specified table. Instead, the full outer join is
executed as a union of left outer join and an antijoin.
9.3.3 Semijoins
A semijoin is a join between two data sets that returns a row from the first set when a
matching row exists in the subquery data set.
The database stops processing the second data set at the first match. Thus,
optimization does not duplicate rows from the first data set when multiple rows in the
second data set satisfy the subquery criteria.
9-39
Chapter 9
Join Types
Note:
Semijoins and antijoins are considered join types even though the SQL
constructs that cause them are subqueries. They are internal algorithms that
the optimizer uses to flatten subquery constructs so that they can be
resolved in a join-like way.
9-40
Chapter 9
Join Types
In the preceding pseudocode, ds1 is the first data set, and ds2_subquery is the
subquery data set. The code obtains the first row from the first data set, and then
loops through the subquery data set looking for a match. The code exits the inner loop
as soon as it finds a match, and then begins processing the next row in the first data
set.
Example 9-13 Semijoin Using WHERE EXISTS
The following query uses a WHERE EXISTS clause to list only the departments that
contain employees:
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes|Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 2
(100)| |
| 1 | NESTED LOOPS SEMI | | 11 | 209 | 2 (0)|
00:00:01 |
| 2 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
|*3 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 44 | 132 | 0
(0)| |
---------------------------------------------------------------------------
-----
For each row in departments, which forms the outer loop, the database obtains the
department ID, and then probes the employees.department_id index for matching
entries. Conceptually, the index looks as follows:
10,rowid
10,rowid
10,rowid
10,rowid
30,rowid
30,rowid
30,rowid
...
If the first entry in the departments table is department 30, then the database performs
a range scan of the index until it finds the first 30 entry, at which point it stops reading
the index and returns the matching row from departments. If the next row in the outer
loop is department 20, then the database scans the index for a 20 entry, and not
9-41
Chapter 9
Join Types
finding any matches, performs the next iteration of the outer loop. The database
proceeds in this way until all matching rows are returned.
Example 9-14 Semijoin Using IN
The following query uses a IN clause to list only the departments that contain
employees:
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes|Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 2
(100)| |
| 1 | NESTED LOOPS SEMI | | 11 | 209 | 2 (0)|
00:00:01 |
| 2 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
|*3 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 44 | 132 | 0
(0)| |
---------------------------------------------------------------------------
-----
9.3.4 Antijoins
An antijoin is a join between two data sets that returns a row from the first set when a
matching row does not exist in the subquery data set.
Like a semijoin, an antijoin stops processing the subquery data set when the first
match is found. Unlike a semijoin, the antijoin only returns a row when no match is
found.
This section contains the following topics:
• When the Optimizer Considers Antijoins
An antijoin avoids unnecessary processing when a query only needs to return a
row when a match does not exist.
• How Antijoins Work
The antijoin optimization is implemented differently depending on what type of join
is used.
9-42
Chapter 9
Join Types
Execution Plan
----------------------------------------------------------
Plan hash value: 1543991079
------------------------------------------------------------------------
---
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | | 14 | 1400 | 5 (20)|
00:00:01 |
|* 1 | HASH JOIN ANTI | | 14 | 1400 | 5 (20)|
00:00:01 |
| 2 | TABLE ACCESS FULL| EMP | 14 | 1218 | 2 (0)|
00:00:01 |
| 3 | TABLE ACCESS FULL| DEPT | 4 | 52 | 2 (0)|
00:00:01 |
------------------------------------------------------------------------
---
1 - access("EMP"."DEPTNO"="DEPT"."DEPTNO")
9-43
Chapter 9
Join Types
Note
-----
- dynamic statistics used: dynamic sampling (level=2)
In the preceding pseudocode, ds1 is the first data set, and ds2 is the second data set.
The code obtains the first row from the first data set, and then loops through the
second data set looking for a match. The code exits the inner loop as soon as it finds a
match, and begins processing the next row in the first data set.
Example 9-15 Semijoin Using WHERE EXISTS
The following query uses a WHERE EXISTS clause to list only the departments that
contain employees:
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes |Cost(%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 2
(100)| |
| 1 | NESTED LOOPS SEMI | | 11 | 209 | 2 (0)|
00:00:01 |
| 2 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
9-44
Chapter 9
Join Types
For each row in departments, which forms the outer loop, the database obtains the
department ID, and then probes the employees.department_id index for matching
entries. Conceptually, the index looks as follows:
10,rowid
10,rowid
10,rowid
10,rowid
30,rowid
30,rowid
30,rowid
...
If the first record in the departments table is department 30, then the database
performs a range scan of the index until it finds the first 30 entry, at which point it stops
reading the index and returns the matching row from departments. If the next row in
the outer loop is department 20, then the database scans the index for a 20 entry, and
not finding any matches, performs the next iteration of the outer loop. The database
proceeds in this way until all matching rows are returned.
If a null value is returned to a NOT IN operator, then the statement returns no records.
To see why, consider the following WHERE clause:
For the entire expression to be true, each individual condition must be true. However,
a null value cannot be compared to another value, so the department_id !=null
condition cannot be true, and thus the whole expression is always false. The
following techniques enable a statement to return records even when nulls are
returned to the NOT IN operator:
9-45
Chapter 9
Join Types
In contrast to NOT IN, the NOT EXISTS clause only considers predicates that return the
existence of a match, and ignores any row that does not match or could not be
determined because of nulls. If at least one row in the subquery matches the row from
the outer query, then NOT EXISTS returns false. If no tuples match, then NOT EXISTS
returns true. The presence of nulls in the subquery does not affect the search for
matching records.
In releases earlier than Oracle Database 11g, the optimizer could not use an antijoin
optimization when nulls could be returned by a subquery. However, starting in Oracle
Database 11g, the ANTI NA (and ANTI SNA) optimizations described in the following
sections enable the optimizer to use an antijoin even when nulls are possible.
Example 9-16 Antijoin Using NOT IN
Suppose that a user issues the following query with a NOT IN clause to list the
departments that contain no employees:
The preceding query returns no rows even though several departments contain no
employees. This result, which was not intended by the user, occurs because the
employees.department_id column is nullable.
The execution plan reveals a NESTED LOOPS ANTI SNA operation in Step 2:
---------------------------------------------------------------------------
-----
| Id| Operation | Name |Rows|Bytes|Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | |
4(100)| |
|*1 | FILTER | | | |
| |
| 2 | NESTED LOOPS ANTI SNA| | 17 | 323 | 4 (50)|
00:00:01 |
| 3 | TABLE ACCESS FULL | DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
|*4 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 41 | 123 | 0
(0)| |
|*5 | TABLE ACCESS FULL | EMPLOYEES | 1 | 3 | 2 (0)|
00:00:01 |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
9-46
Chapter 9
Join Types
---------------------------------------------------
1 - filter( IS NULL)
4 - access("DEPARTMENT_ID"="DEPARTMENT_ID")
5 - filter("DEPARTMENT_ID" IS NULL)
The ANTI SNA stands for "single null-aware antijoin." ANTI NA stands for "null-aware
antijoin." The null-aware operation enables the optimizer to use the semijoin
optimization even on a nullable column. In releases earlier than Oracle Database 11g,
the database could not perform antijoins on NOT IN queries when nulls were possible.
Suppose that the user rewrites the query by applying an IS NOT NULL condition to the
subquery:
The preceding query returns 16 rows, which is the expected result. Step 1 in the plan
shows a standard NESTED LOOPS ANTI join instead of an ANTI NA or ANTI SNA join
because the subquery cannot returns nulls:
---------------------------------------------------------------------------
-----
|Id| Operation | Name | Rows| Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0| SELECT STATEMENT | | | | 2
(100)| |
| 1| NESTED LOOPS ANTI | | 17 | 323 | 2 (0)|
00:00:01 |
| 2| TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01 |
|*3| INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 41 | 123 | 0
(0)| |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("DEPARTMENT_ID"="DEPARTMENT_ID")
filter("DEPARTMENT_ID" IS NOT NULL)
9-47
Chapter 9
Join Types
The preceding query avoids the null problem for NOT IN clauses. Thus, even though
employees.department_id column is nullable, the statement returns the desired result.
Step 1 of the execution plan reveals a NESTED LOOPS ANTI operation, not the ANTI NA
variant, which was necessary for NOT IN when nulls were possible:
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows | Bytes | Cost (%CPU)|
Time|
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 2
(100)| |
| 1 | NESTED LOOPS ANTI | | 17 | 323 | 2 (0)|
00:00:01|
| 2 | TABLE ACCESS FULL| DEPARTMENTS | 27 | 432 | 2 (0)|
00:00:01|
|*3 | INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 41 | 123 | 0
(0)| |
---------------------------------------------------------------------------
-----
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("E"."DEPARTMENT_ID"="D"."DEPARTMENT_ID")
9-48
Chapter 9
Join Types
rows resulting from the join is calculated using the following formula, where rs1 is the
number of rows in first row set and rs2 is the number of rows in the second row set:
Note:
If a Cartesian join appears in a query plan, it could be caused by an
inadvertently omitted join condition. In general, if a query joins n tables,
then n-1 join conditions are required to avoid a Cartesian join.
At a high level, the algorithm for a Cartesian join looks as follows, where ds1 is
typically the smaller data set, and ds2 is the larger data set:
9-49
Chapter 9
Join Types
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows | Bytes |Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | |11
(100)| |
| 1 | MERGE JOIN CARTESIAN | | 2889 | 57780 |11 (0)|
00:00:01 |
| 2 | TABLE ACCESS FULL | DEPARTMENTS | 27 | 324 | 2 (0)|
00:00:01 |
| 3 | BUFFER SORT | | 107 | 856 | 9 (0)|
00:00:01 |
| 4 | INDEX FAST FULL SCAN| EMP_NAME_IX | 107 | 856 | 0
(0)| |
---------------------------------------------------------------------------
-----
In Step 1 of the preceding plan, the CARTESIAN keyword indicates the presence of a
Cartesian join. The number of rows (2889) is the product of 27 and 107.
In Step 3, the BUFFER SORT operation indicates that the database is copying the data
blocks obtained by the scan of emp_name_ix from the SGA to the PGA. This strategy
avoids multiple scans of the same blocks in the database buffer cache, which would
generate many logical reads and permit resource contention.
9-50
Chapter 9
Join Optimizations
The following execution plan shows a Cartesian product (Step 3) between locations
(Step 6) and employees (Step 4), which is then joined to the departments table (Step
2):
---------------------------------------------------------------------------
-----
| Id| Operation | Name | Rows | Bytes |Cost (%CPU)|
Time |
---------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | | | 37
(100)| |
|*1 | HASH JOIN | | 106 | 4664 | 37 (6)|
00:00:01 |
| 2 | TABLE ACCESS FULL | DEPARTMENTS | 27 | 513 | 2 (0)|
00:00:01 |
| 3 | MERGE JOIN CARTESIAN| | 2461 | 61525 | 34 (3)|
00:00:01 |
| 4 | TABLE ACCESS FULL | EMPLOYEES | 107 | 1177 | 2 (0)|
00:00:01 |
| 5 | BUFFER SORT | | 23 | 322 | 32 (4)|
00:00:01 |
| 6 | TABLE ACCESS FULL | LOCATIONS | 23 | 322 | 0
(0)| |
---------------------------------------------------------------------------
-----
See Also:
Oracle Database SQL Language Reference to learn about the ORDERED hint
9-51
Chapter 9
Join Optimizations
9-52
Chapter 9
Join Optimizations
For example, 8 elements (an arbitrary number used for this example) in an array are
initially set to 0:
e1 e2 e3 e4 e5 e6 e7 e8
0 0 0 0 0 0 0 0
This array represents a set. To represent an input value i in this array, three separate
hash functions (three is arbitrary) are applied to i, each generating a hash value
between 1 and 8:
f1(i) = h1
f2(i) = h2
f3(i) = h3
For example, to store the value 17 in this array, the hash functions set i to 17, and then
return the following hash values:
f1(17) = 5
f2(17) = 3
f3(17) = 5
In the preceding example, two of the hash functions happened to return the same
value of 5, known as a hash collision. Because the distinct hash values are 5 and 3,
the 5th and 3rd elements in the array are set to 1:
e1 e2 e3 e4 e5 e6 e7 e8
0 0 1 0 1 0 0 0
Testing the membership of 17 in the set reverses the process. To test whether the set
excludes the value 17, element 3 or element 5 must contain a 0. If a 0 is present in
either element, then the set cannot contain 17. No false negatives are possible.
To test whether the set includes 17, both element 3 and element 5 must contain 1
values. However, if the test indicates a 1 for both elements, then it is still possible for
the set not to include 17. False positives are possible. For example, the following array
might represent the value 22, which also has a 1 for both element 3 and element 5:
e1 e2 e3 e4 e5 e6 e7 e8
1 0 1 0 1 0 0 0
9-53
Chapter 9
Join Optimizations
See Also:
Oracle Database SQL Language Reference to learn more about the bloom
filter hints
---------------------------------------------------------------------------
-
| Id | Operation | Name | TQ |IN-OUT| PQ Distrib
|
---------------------------------------------------------------------------
-
...
| 9 | JOIN FILTER CREATE | :BF0000 | Q1,03 | PCWP |
|
In the Predicate Information section of the plan, filters that contain functions
beginning with the string SYS_OP_BLOOM_FILTER indicate use of a Bloom filter.
9-54
Chapter 9
Join Optimizations
SELECT * FROM
TABLE(DBMS_XPLAN.DISPLAY_CURSOR(format => 'BASIC,+PARALLEL,+PREDICATE'));
---------------------------------------------------------------------------
-
| Id | Operation | Name | TQ |IN-OUT| PQ Distrib
|
---------------------------------------------------------------------------
-
| 0 | SELECT STATEMENT | | | |
|
| 1 | PX COORDINATOR | | | |
|
| 2 | PX SEND QC (RANDOM) | :TQ10003 | Q1,03 | P->S | QC (RAND)
|
|* 3 | HASH JOIN BUFFERED | | Q1,03 | PCWP |
|
| 4 | PX RECEIVE | | Q1,03 | PCWP |
|
| 5 | PX SEND BROADCAST | :TQ10001 | Q1,01 | S->P | BROADCAST
|
| 6 | PX SELECTOR | | Q1,01 | SCWC |
|
| 7 | TABLE ACCESS FULL | PRODUCTS | Q1,01 | SCWP |
|
|* 8 | HASH JOIN | | Q1,03 | PCWP |
|
| 9 | JOIN FILTER CREATE | :BF0000 | Q1,03 | PCWP |
|
| 10 | BUFFER SORT | | Q1,03 | PCWC |
|
| 11 | PX RECEIVE | | Q1,03 | PCWP |
|
| 12 | PX SEND HYBRID HASH| :TQ10000 | | S->P | HYBRID
HASH|
|* 13 | TABLE ACCESS FULL | TIMES | | |
|
| 14 | PX RECEIVE | | Q1,03 | PCWP |
|
| 15 | PX SEND HYBRID HASH | :TQ10002 | Q1,02 | P->P | HYBRID
HASH|
| 16 | JOIN FILTER USE | :BF0000 | Q1,02 | PCWP |
|
| 17 | PX BLOCK ITERATOR | | Q1,02 | PCWC |
|
9-55
Chapter 9
Join Optimizations
3 - access("S"."PROD_ID"="P"."PROD_ID")
8 - access("S"."TIME_ID"="T"."TIME_ID")
13 - filter("T"."FISCAL_WEEK_NUMBER"=18)
18 - access(:Z>=:Z AND :Z<=:Z)
filter(SYS_OP_BLOOM_FILTER(:BF0000,"S"."TIME_ID"))
A single server process scans the times table (Step 13), and then uses a hybrid hash
distribution method to send the rows to the parallel execution servers (Step 12). The
processes in set Q1,03 create a bloom filter (Step 9). The processes in set Q1,02 scan
sales in parallel (Step 18), and then use the Bloom filter to discard rows from sales
(Step 16) before sending them on to set Q1,03 using hybrid hash distribution (Step 15).
The processes in set Q1,03 hash join the times rows to the filtered sales rows (Step
8). The processes in set Q1,01 scan products (Step 7), and then send the rows to
Q1,03 (Step 5). Finally, the processes in Q1,03 join the products rows to the rows
generated by the previous hash join (Step 3).
The following figure illustrates the basic process.
Q1, 03
Create
Bloom filter
:BF0000
Q1, 01 Q1, 02
9-56
Chapter 9
Join Optimizations
Only one table is partitioned on the join key. The other table may or may not be
partitioned.
This section contains the following topics:
See Also:
Oracle Database VLDB and Partitioning Guide explains partition-wise joins in
detail
9-57
Chapter 9
Join Optimizations
Server
Process
t1 t2
9-58
Chapter 9
Join Optimizations
PE Coordinator
t1 t2
PE Server
PE Server
PE Server
PE Server
A full partition-wise join can also join partitions to subpartitions, which is useful when
the tables use different partitioning methods. For example, customers is partitioned by
hash, but sales is partitioned by range. If you subpartition sales by hash, then the
database can perform a full partition-wise join between the hash partitions of the
customers and the hash subpartitions of sales.
In the execution plan, the presence of a partition operation before the join signals the
presence of a full partition-wise join, as in the following snippet:
| 8 | PX PARTITION HASH ALL|
|* 9 | HASH JOIN |
See Also:
Oracle Database VLDB and Partitioning Guide explains full partition-wise
joins in detail, and includes several examples
9-59
Chapter 9
Join Optimizations
PE Coordinator
t1 t2
PE Server
PE Server t1 t2
PE Server
PE Server
PE Server
PE Server
PE Server
PE Server
Dynamically created
partitions
See Also:
Oracle Database VLDB and Partitioning Guide explains full partition-wise
joins in detail, and includes several examples
9-60
Chapter 9
Join Optimizations
See Also:
Oracle Database In-Memory Guide to learn how to optimize In-Memory
queries with join groups
9-61
Part V
Optimizer Statistics
The accuracy of an execution plan depends on the quality of the optimizer statistics.
This part contains the following chapters:
• Optimizer Statistics Concepts
Oracle Database optimizer statistics describe details about the database and its
objects.
• Histograms
A histogram is a special type of column statistic that provides more detailed
information about the data distribution in a table column. A histogram sorts values
into "buckets," as you might sort coins into buckets.
• Configuring Options for Optimizer Statistics Gathering
This chapter explains what optimizer statistics collection is and how to set
statistics preferences.
• Gathering Optimizer Statistics
This chapter explains how to use the DBMS_STATS.GATHER_*_STATS program units.
• Managing Extended Statistics
DBMS_STATS enables you to collect extended statistics, which are statistics that
can improve cardinality estimates when multiple predicates exist on different
columns of a table, or when predicates use expressions.
• Controlling the Use of Optimizer Statistics
Using DBMS_STATS, you can specify when and how the optimizer uses statistics.
• Managing Historical Optimizer Statistics
This chapter how to retain, report on, and restore non-current statistics.
• Importing and Exporting Optimizer Statistics
You can export and import optimizer statistics from the data dictionary to user-
defined statistics tables. You can also copy statistics from one database to another
database.
• Analyzing Statistics Using Optimizer Statistics Advisor
Optimizer Statistics Advisor analyzes how optimizer statistics are gathered, and
then makes recommendations.
10
Optimizer Statistics Concepts
Oracle Database optimizer statistics describe details about the database and its
objects.
This chapter includes the following topics:
• Introduction to Optimizer Statistics
The optimizer cost model relies on statistics collected about the objects involved
in a query, and the database and host where the query runs.
• About Optimizer Statistics Types
The optimizer collects statistics on different types of database objects and
characteristics of the database environment.
• How the Database Gathers Optimizer Statistics
Oracle Database provides several mechanisms to gather statistics.
• When the Database Gathers Optimizer Statistics
The database collects optimizer statistics at various times and from various
sources.
Related Topics
• Query Optimizer Concepts
This chapter describes the most important concepts relating to the query
optimizer, including its principal components.
• Histograms
A histogram is a special type of column statistic that provides more detailed
information about the data distribution in a table column. A histogram sorts values
into "buckets," as you might sort coins into buckets.
• Gathering Optimizer Statistics
This chapter explains how to use the DBMS_STATS.GATHER_*_STATS program units.
• Managing Historical Optimizer Statistics
This chapter how to retain, report on, and restore non-current statistics.
10-1
Chapter 10
Introduction to Optimizer Statistics
– Number of blocks
– Average row length
• Column statistics
– Number of distinct values (NDV) in a column
– Number of nulls in a column
– Data distribution (histogram)
– Extended statistics
• Index statistics
– Number of leaf blocks
– Number of levels
– Index clustering factor
• System statistics
– I/O performance and utilization
– CPU performance and utilization
As shown in Figure 10-1, the database stores optimizer statistics for tables, columns,
indexes, and the system in the data dictionary. You can access these statistics using
data dictionary views.
Note:
The optimizer statistics are different from the performance statistics visible
through V$ views.
10-2
Chapter 10
About Optimizer Statistics Types
Database
Optimizer
Data Dictionary
Optimizer Statistics
PERSON
Table GB Execution
Plan
ID Name HJ
100 Kumar HJ
PERSON_ID_IX
10-3
Chapter 10
About Optimizer Statistics Types
• System Statistics
The system statistics describe hardware characteristics such as I/O and CPU
performance and utilization.
• User-Defined Optimizer Statistics
The extensible optimizer enables authors of user-defined functions and indexes
to create statistics collection, selectivity, and cost functions. The optimizer cost
model is extended to integrate information supplied by the user to assess CPU
and the I/O cost.
See Also:
10-4
Chapter 10
About Optimizer Statistics Types
See Also:
• "Histograms "
• "About Statistics on Column Groups"
• Oracle Database Reference for a description of the
DBA_TAB_COL_STATISTICS view
10-5
Chapter 10
About Optimizer Statistics Types
The BLEVEL column shows the number of blocks required to go from the root block
to a leaf block. A B-tree index has two types of blocks: branch blocks for searching
and leaf blocks that store values. See Oracle Database Concepts for a conceptual
overview of B-tree indexes.
• Distinct keys
This columns tracks the number of distinct indexed values. If a unique constraint is
defined, and if no NOT NULL constraint is defined, then this value equals the
number of non-null values.
• Average number of leaf blocks for each distinct indexed key
• Average number of data blocks pointed to by each distinct indexed key
See Also:
Oracle Database Reference for a description of the DBA_IND_STATISTICS
view
10-6
Chapter 10
About Optimizer Statistics Types
The clustering factor is a property of a specific index, not a table. If multiple indexes
exist on a table, then the clustering factor for one index might be small while the factor
for another index is large. An attempt to reorganize the table to improve the clustering
factor for one index may degrade the clustering factor of the other index.
Example 10-3 Index Clustering Factor
This example shows how the optimizer uses the index clustering factor to determine
whether using an index is more effective than a full table scan.
1. Start SQL*Plus and connect to a database as sh, and then query the number of
rows and blocks in the sh.customers table (sample output included):
4. Create a new copy of the customers table, with rows ordered by cust_last_name.
For example, execute the following statements:
10-7
Chapter 10
About Optimizer Statistics Types
The table customers3 has the same data as the original customers table, but the
index on customers3 has a much lower clustering factor because the data in the
table is ordered by the cust_last_name. The clustering factor is now about 10
times the number of blocks instead of 70 times.
9. Query the customers table.
For example, execute the following query (sample output included):
CUST_FIRST_NAME CUST_LAST_NAME
-------------------- ----------------------------------------
Vida Puleo
Harriett Quinlan
Madeleine Quinn
Caresse Puleo
10-8
Chapter 10
About Optimizer Statistics Types
For example, execute the following query (partial sample output included):
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows |Bytes|Cost (%CPU)| Time |
-------------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | | 405 (100)| |
|* 1| TABLE ACCESS STORAGE FULL| CUSTOMERS | 2335|35025| 405 (1)|00:00:01|
-------------------------------------------------------------------------------
The preceding plan shows that the optimizer did not use the index on the original
customers tables.
11. Query the customers3 table.
CUST_FIRST_NAME CUST_LAST_NAME
-------------------- ----------------------------------------
Vida Puleo
Harriett Quinlan
Madeleine Quinn
Caresse Puleo
For example, execute the following query (partial sample output included):
---------------------------------------------------------------------------------------
|Id| Operation | Name |Rows|Bytes|Cost(%CPU)| Time|
---------------------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | |69(100)| |
| 1| TABLE ACCESS BY INDEX ROWID|CUSTOMERS3 |2335|35025|69(0) |00:00:01|
|*2| INDEX RANGE SCAN |CUSTOMERS3_LAST_NAME_IDX|2335| |7(0) |00:00:01|
---------------------------------------------------------------------------------------
The result set is the same, but the optimizer chooses the index. The plan cost is
much less than the cost of the plan used on the original customers table.
13. Query customers with a hint that forces the optimizer to use the index.
For example, execute the following query (partial sample output included):
10-9
Chapter 10
About Optimizer Statistics Types
CUST_FIRST_NAME CUST_LAST_NAME
-------------------- ----------------------------------------
Vida Puleo
Caresse Puleo
Harriett Quinlan
Madeleine Quinn
For example, execute the following query (partial sample output included):
---------------------------------------------------------------------------------------
--
| Id | Operation | Name |Rows|Bytes|Cost(%CPU)|
Time |
---------------------------------------------------------------------------------------
--
| 0| SELECT STATEMENT | | | | 422(100)
| |
| 1| TABLE ACCESS BY INDEX ROWID|CUSTOMERS |335 |35025| 422(0) |
00:00:01|
|*2| INDEX RANGE SCAN |CUSTOMERS_LAST_NAME_IDX|2335| | 7(0) |
00:00:01|
---------------------------------------------------------------------------------------
--
The preceding plan shows that the cost of using the index on customers is higher
than the cost of a full table scan. Thus, using an index does not necessarily
improve performance. The index clustering factor is a measure of whether an
index scan is more effective than a full table scan.
In this example, the index clustering factor for col1_idx is low. The rows that have the
same indexed column values for col1 are in the same data blocks in the table. Thus,
10-10
Chapter 10
About Optimizer Statistics Types
the cost of using an index range scan to return all rows with value A is low because
only one block in the table must be read.
Example 10-5 Scattered Data
Assume that the same rows are scattered across the data blocks as follows:
In this example, the index clustering factor for col1_idx is higher. The database must
read all three blocks in the table to retrieve all rows with the value A in col1.
See Also:
Oracle Database Reference for a description of the DBA_INDEXES view
When set to session-specific, you can gather statistics for a global temporary table in
one session, and then use the statistics for this session only. Meanwhile, users can
continue to maintain a shared version of the statistics. During optimization, the
10-11
Chapter 10
About Optimizer Statistics Types
optimizer first checks whether a global temporary table has session-specific statistics.
If yes, the optimizer uses them. Otherwise, the optimizer uses shared statistics if they
exist.
Session-specific statistics have the following characteristics:
• Dictionary views that track statistics show both the shared statistics and the
session-specific statistics in the current session.
The views are DBA_TAB_STATISTICS, DBA_IND_STATISTICS, DBA_TAB_HISTOGRAMS,
and DBA_TAB_COL_STATISTICS (each view has a corresponding USER_ and ALL_
version). The SCOPE column shows whether statistics are session-specific or
shared.
• Other sessions do not share the cursor using the session-specific statistics.
Different sessions can share the cursor using shared statistics, as in releases
earlier than Oracle Database 12c. The same session can share the cursor using
session-specific statistics.
• Pending statistics are not supported for session-specific statistics.
• When the GLOBAL_TEMP_TABLE_STATS preference is set to SESSION, by default
GATHER_TABLE_STATS immediately invalidates previous cursors compiled in the
same session. However, this procedure does not invalidate cursors compiled in
other sessions.
10-12
Chapter 10
About Optimizer Statistics Types
the shared statistics, and GATHER_TABLE_STATS deletes all rows from a transaction-
specific temporary table.
See Also:
See Also:
10-13
Chapter 10
How the Database Gathers Optimizer Statistics
See Also:
This package enables you to control what and how statistics are collected, including
the degree of parallelism for statistics collection, sampling methods, granularity of
statistics collection in partitioned tables, and so on.
Note:
Do not use the COMPUTE and ESTIMATE clauses of the ANALYZE statement to
collect optimizer statistics. These clauses have been deprecated. Instead,
use DBMS_STATS.
10-14
Chapter 10
How the Database Gathers Optimizer Statistics
Statistics gathered with the DBMS_STATS package are required for the creation of
accurate execution plans. For example, table statistics gathered by DBMS_STATS
include the number of rows, number of blocks, and average row length.
By default, Oracle Database uses automatic optimizer statistics collection. In this case,
the database automatically runs DBMS_STATS to collect optimizer statistics for all
schema objects for which statistics are missing or stale. The process eliminates many
manual tasks associated with managing the optimizer, and significantly reduces the
risks of generating suboptimal execution plans because of missing or stale statistics.
You can also update and manage optimizer statistics by manually executing
DBMS_STATS.
See Also:
Note:
Dynamic statistics augment statistics rather than providing an alternative to
them.
Dynamic statistics supplement optimizer statistics such as table and index block
counts, table and join cardinalities (estimated number of rows), join column statistics,
and GROUP BY statistics. This information helps the optimizer improve plans by making
better estimates for predicate cardinality.
Dynamic statistics are beneficial in the following situations:
• An execution plan is suboptimal because of complex predicates.
• The sampling time is a small fraction of total execution time for the query.
• The query executes many times so that the sampling time is amortized.
10-15
Chapter 10
How the Database Gathers Optimizer Statistics
Related Topics
• When the Database Samples Data
Starting in Oracle Database 12c, the optimizer automatically decides whether
dynamic statistics are useful and which sample size to use for all SQL statements.
In earlier releases, dynamic statistics were called dynamic sampling.
• Guideline for Setting the Sample Size
In the context of optimizer statistics, sampling is the gathering of statistics from a
random subset of table rows. By enabling the database to avoid full table scans
and sorts of entire tables, sampling minimizes the resources necessary to gather
statistics.
• Configuring Options for Dynamic Statistics
Dynamic statistics are an optimization technique in which the database uses
recursive SQL to scan a small random sample of the blocks in a table.
Note:
By default, a parallel insert uses a direct path insert. You can force a direct
path insert by using the /*+APPEND*/ hint.
10-16
Chapter 10
How the Database Gathers Optimizer Statistics
See Also:
Oracle Database Data Warehousing Guide to learn more about bulk loads
See Also:
10-17
Chapter 10
How the Database Gathers Optimizer Statistics
The preceding example only gathers missing or stale statistics. The database does not
gather table and basic column statistics collected during the bulk load.
Note:
You can set the table preference options to GATHER AUTO on the tables that
you plan to bulk load. In this way, you need not explicitly set the options
parameter when running GATHER_TABLE_STATS.
See Also:
10-18
Chapter 10
How the Database Gathers Optimizer Statistics
Note:
The DBA_TAB_COL_STATISTICS.NOTES column is set to STATS_ON_LOAD by
a bulk load into an empty table. However, subsequent bulk loads into the
non-empty table do not reset the NOTES column. One technique for
determining whether the database gathered statistics is to query the
USER_TAB_MODIFICATIONS.INSERTS column. If the query returns a row
indicating the number of rows loaded, then the most recent bulk load did
not gather statistics automatically.
• It is loaded using an INSERT INTO ... SELECT, and neither of the following
conditions is true: all columns of the target table are specified, or a subset of the
target columns are specified and the unspecified columns have default values.
Put differently, the database only gathers statistics automatically for bulk loads
when either all columns of the target table are specified, or a subset of the target
columns are specified and the unspecified columns have default values. For
example, the sales table has only columns c1, c2, c3, and c4. The column c4 does
not have a default value. You load sales_copy by executing INSERT /*+ APPEND
*/ INTO sales_copy SELECT c1, c2, c3 FROM sales. In this case, the database
does not gather online statistics for sales_copy. The database would gather
statistics if c4 had a default value or if it were included in the SELECT list.
• It is in an Oracle-owned schema such as SYS.
• It is one of the following types of tables: nested table, index-organized table (IOT),
external table, or global temporary table defined as ON COMMIT DELETE ROWS.
• It has a PUBLISH preference set to FALSE.
• Its statistics are locked.
• It is partitioned, INCREMENTAL is set to true, and partition-extended syntax is not
used.
For example, assume that you execute DBMS_STATS.SET_TABLE_PREFS(null,
'sales', incremental', 'true'). In this case, the database does not gather
statistics for INSERT INTO sales SELECT, even when sales is empty. However,
the database does gather statistics automatically for INSERT INTO sales
PARTITION (sales_q4_2000) SELECT.
• It is loaded using a multitable INSERT statement.
See Also:
10-19
Chapter 10
When the Database Gathers Optimizer Statistics
For example, the following statement disables online statistics gathering for bulk loads:
See Also:
Oracle Database SQL Language Reference to learn about the
GATHER_OPTIMIZER_STATISTICS and NO_GATHER_OPTIMIZER_STATISTICS hints
10-20
Chapter 10
When the Database Gathers Optimizer Statistics
During SQL compilation, the database can augment the statistics previously
gathered by DBMS_STATS. In this stage, the database runs additional queries to
obtain more accurate information on how many rows in the tables satisfy the WHERE
clause predicates in the SQL statement.
• SQL execution
During execution, the database can further augment previously gathered statistics.
In this stage, Oracle Database collects the number of rows produced by every row
source during the execution of a SQL statement. At the end of execution, the
optimizer determines whether the estimated number of rows is inaccurate enough
to warrant reparsing at the next statement execution. If the cursor is marked for
reparsing, then the optimizer uses actual row counts from the previous execution
instead of estimates.
• SQL profiles
A SQL profile is a collection of auxiliary statistics on a query. The profile stores
these supplemental statistics in the data dictionary. The optimizer uses SQL
profiles during optimization to determine the most optimal plan.
The database stores optimizer statistics in the data dictionary and updates or replaces
them as needed. You can query statistics in data dictionary views.
See Also:
10-21
Chapter 10
When the Database Gathers Optimizer Statistics
See Also:
Oracle Database Reference to learn more about DBA_SQL_PLAN_DIRECTIVES
10-22
Chapter 10
When the Database Gathers Optimizer Statistics
Dynamic statistics have some performance overhead. Every time the optimizer
hard parses a query to which a dynamic statistics directive applies, the database
must perform the extra sampling.
Starting in Oracle Database 12c Release 2 (12.2), the database writes statistics
from adaptive dynamic sampling to the SQL plan directives store, making them
available to other queries.
• Column groups
The optimizer examines the query corresponding to the directive. If there is a
missing column group, and if the DBMS_STATS preference AUTO_STAT_EXTENSIONS is
set to ON (the default is OFF) for the affected table, then the optimizer automatically
creates this column group the next time DBMS_STATS gathers statistics on the table.
Otherwise, the optimizer does not automatically create the column group.
If a column group exists, then the next time this statement executes, the optimizer
uses the column group statistics in place of the SQL plan directive when possible
(equality predicates, GROUP BY, and so on). In subsequent executions, the
optimizer may create additional SQL plan directives to address other problems in
the plan, such as join or GROUP BY cardinality misestimates.
Note:
Currently, the optimizer monitors only column groups. The optimizer
does not create an extension on expressions.
When the problem that occasioned a directive is solved, either because a better
directive exists or because a histogram or extension exists, the
DBA_SQL_PLAN_DIRECTIVES.STATE value changes from USABLE to SUPERSEDED. More
information about the directive state is exposed in the
DBA_SQL_PLAN_DIRECTIVES.NOTES column.
See Also:
10-23
Chapter 10
When the Database Gathers Optimizer Statistics
You can manage directives by using the DBMS_SPD package. For example, you can:
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_SPD package
Assumptions
You plan to run queries against the sh schema, and you have privileges on this
schema and on data dictionary and V$ views.
SELECT /*+gather_plan_statistics*/ *
FROM customers
WHERE cust_state_province='CA'
AND country_id='US';
The gather_plan_statistics hint shows the actual number of rows returned from
each operation in the plan. Thus, you can compare the optimizer estimates with
the actual number of rows returned.
2. Query the plan for the preceding query.
The following example shows the execution plan (sample output included):
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID b74nw722wjvy3, child number 0
-------------------------------------
select /*+gather_plan_statistics*/ * from customers where
CUST_STATE_PROVINCE='CA' and country_id='US'
10-24
Chapter 10
When the Database Gathers Optimizer Statistics
---------------------------------------------------------------------------------------
-----
| Id| Operation | Name | Starts |E-Rows| A-Rows | A-Time |Buffers |
Reads |
---------------------------------------------------------------------------------------
-----
| 0 | SELECT STATEMENT | | 1 | | 29 |00:00:00.01 | 17
| 14 |
|*1 | TABLE ACCESS FULL| CUSTOMERS | 1 | 8 | 29 |00:00:00.01 | 17
| 14 |
---------------------------------------------------------------------------------------
-----
The actual number of rows (A-Rows) returned by each operation in the plan varies
greatly from the estimates (E-Rows). This statement is a candidate for automatic
reoptimization.
3. Check whether the customers query can be reoptimized.
The following statement queries the V$SQL.IS_REOPTIMIZABLE value (sample
output included):
10-25
Chapter 10
When the Database Gathers Optimizer Statistics
The following example uses DBMS_SPD to write the SQL plan directives to disk, and
then shows the directives for the sh schema only:
EXEC DBMS_SPD.FLUSH_SQL_PLAN_DIRECTIVE;
CARDINALITY
MISESTIMATE
1484026771529551585 SH CUSTOMERS CUST_STATE_ COLUMN DYNAMIC_SAMPLING USABLE SINGLE
TABLE
PROVINCE
CARDINALITY
MISESTIMATE
1484026771529551585 SH CUSTOMERS TABLE DYNAMIC_SAMPLING USABLE SINGLE
TABLE
CARDINALITY
MISESTIMATE
Initially, the database stores SQL plan directives in memory, and then writes them
to disk every 15 minutes. Thus, the preceding example calls
DBMS_SPD.FLUSH_SQL_PLAN_DIRECTIVE to force the database to write the directives
to the SYSAUX tablespace.
Monitor directives using the views DBA_SQL_PLAN_DIRECTIVES and
DBA_SQL_PLAN_DIR_OBJECTS. Three entries appear in the views, one for the
customers table itself, and one for each of the correlated columns. Because the
customers query has the IS_REOPTIMIZABLE value of Y, if you reexecute the
statement, then the database will hard parse it again, and then generate a plan
based on the previous execution statistics.
5. Query the customers table again.
For example, enter the following statement:
SELECT /*+gather_plan_statistics*/ *
FROM customers
WHERE cust_state_province='CA'
AND country_id='US';
10-26
Chapter 10
When the Database Gathers Optimizer Statistics
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID b74nw722wjvy3, child number 1
-------------------------------------
select /*+gather_plan_statistics*/ * from customers where
CUST_STATE_PROVINCE='CA' and country_id='US'
Note
-----
- cardinality feedback used for this statement
The Note section indicates that the database used reoptimization for this
statement. The estimated number of rows (E-Rows) is now correct. The SQL plan
directive has not been used yet.
7. Query the cursors for the customers query.
For example, run the following query (sample output included):
10-27
Chapter 10
When the Database Gathers Optimizer Statistics
TE_PROVINCE
='CA' and c
ountry_id='
US'
A new plan exists for the customers query, and also a new child cursor.
8. Confirm that a SQL plan directive exists and is usable for other statements.
For example, run the following query, which is similar but not identical to the
original customers query (the state is MA instead of CA):
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID 3tk6hj3nkcs2u, child number 0
-------------------------------------
Select /*+gather_plan_statistics*/ cust_email From customers Where
cust_state_province='MA' And country_id='US'
------------------------------------------------------------------------
---
|Id | Operation | Name |Starts|E-Rows|A-Rows| A-Time |
Buffers|
------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | | 1 | | 2 |00:00:00.01|
16 |
|*1 | TABLE ACCESS FULL| CUSTOMERS | 1 | 2 | 2 |00:00:00.01|
16 |
------------------------------------------------------------------------
---
10-28
Chapter 10
When the Database Gathers Optimizer Statistics
Note
-----
- dynamic sampling used for this statement (level=2)
- 1 Sql Plan Directive used for this statement
The Note section of the plan shows that the optimizer used the SQL directive for
this statement, and also used dynamic statistics.
See Also:
• "Automatic Reoptimization"
• "Managing SQL Plan Directives"
• Oracle Database Reference to learn about DBA_SQL_PLAN_DIRECTIVES,
V$SQL, and other database views
• Oracle Database Reference to learn about DBMS_SPD
10.4.2.5 How the Optimizer Uses Extensions and SQL Plan Directives:
Example
The example shows how the database uses a SQL plan directive until the optimizer
verifies that an extension exists and the statistics are applicable.
At this point, the directive changes its status to SUPERSEDED. Subsequent compilations
use the statistics instead of the directive.
Assumptions
This example assumes you have already followed the steps in "How the Optimizer
Uses SQL Plan Directives: Example".
To see how the optimizer uses an extension and SQL plan directive:
1. Gather statistics for the sh.customers table.
For example, execute the following PL/SQL program:
BEGIN
DBMS_STATS.GATHER_TABLE_STATS('SH','CUSTOMERS');
END;
/
10-29
Chapter 10
When the Database Gathers Optimizer Statistics
The preceding output indicates that a column group extension exists on the
cust_state_province and country_id columns.
3. Query the state of the SQL plan directive.
Example 10-6 queries the data dictionary for information about the directive.
Although column group statistics exist, the directive has a state of USABLE because
the database has not yet recompiled the statement. During the next compilation,
the optimizer verifies that the statistics are applicable. If they are applicable, then
the status of the directive changes to SUPERSEDED. Subsequent compilations use
the statistics instead of the directive.
4. Query the sh.customers table.
SELECT /*+gather_plan_statistics*/ *
FROM customers
WHERE cust_state_province='CA'
AND country_id='US';
10-30
Chapter 10
When the Database Gathers Optimizer Statistics
If the cursor is in the shared SQL area, then the database typically shares the
cursor. To force a reparse, this step changes the SQL text slightly by adding a
comment.
8. Query the plan in the cursor.
Example 10-9 shows the execution plan (sample output included).
The absence of a Note shows that the optimizer used the extended statistics
instead of the SQL plan directive. If the directive is not used for 53 weeks, then the
database automatically purges it.
See Also:
EXEC DBMS_SPD.FLUSH_SQL_PLAN_DIRECTIVE;
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID b74nw722wjvy3, child number 0
-------------------------------------
select /*+gather_plan_statistics*/ * from customers where
CUST_STATE_PROVINCE='CA' and country_id='US'
10-31
Chapter 10
When the Database Gathers Optimizer Statistics
-----------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
-----------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 29 |00:00:00.01 | 16 |
|* 1 | TABLE ACCESS FULL| CUSTOMERS | 1 | 29 | 29 |00:00:00.01 | 16 |
-----------------------------------------------------------------------------------------
Note
-----
- dynamic sampling used for this statement (level=2)
- 1 Sql Plan Directive used for this statement
EXEC DBMS_SPD.FLUSH_SQL_PLAN_DIRECTIVE;
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID b74nw722wjvy3, child number 0
-------------------------------------
select /*+gather_plan_statistics*/ * from customers where
CUST_STATE_PROVINCE='CA' and country_id='US'
10-32
Chapter 10
When the Database Gathers Optimizer Statistics
-----------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
-----------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 29 |00:00:00.01 | 17 |
|* 1 | TABLE ACCESS FULL| CUSTOMERS | 1 | 29 | 29 |00:00:00.01 | 17 |
-----------------------------------------------------------------------------------------
19 rows selected.
In general, the optimizer uses default statistics rather than dynamic statistics to
compute statistics needed during optimizations on tables, indexes, and columns. The
optimizer decides whether to use dynamic statistics based on several factors,
including the following:
• The SQL statement uses parallel execution.
• A SQL plan directive exists.
10-33
Chapter 10
When the Database Gathers Optimizer Statistics
Optimizer
GB Execution
Plan
Statistics missing? HJ
Statistics insufficient? No HJ
SQL directive exists?
Parallel execution?
Yes
CLIENT
SQL
Determine sampling Pass results to SELECT ...
size optimizer for FROM sales
use in plan WHERE ...
generation
Sales
As shown in Figure 10-2, the optimizer automatically gathers dynamic statistics in the
following cases:
• Missing statistics
When tables in a query have no statistics, the optimizer gathers basic statistics on
these tables before optimization. Statistics can be missing because the application
creates new objects without a follow-up call to DBMS_STATS to gather statistics, or
because statistics were locked on an object before statistics were gathered.
In this case, the statistics are not as high-quality or as complete as the statistics
gathered using the DBMS_STATS package. This trade-off is made to limit the impact
on the compile time of the statement.
• Insufficient statistics
Statistics can be insufficient whenever the optimizer estimates the selectivity of
predicates (filter or join) or the GROUP BY clause without taking into account
correlation between columns, skew in the column data distribution, statistics on
expressions, and so on.
Extended statistics help the optimizer obtain accurate quality cardinality estimates
for complex predicate expressions. The optimizer can use dynamic statistics to
compensate for the lack of extended statistics or when it cannot use extended
statistics, for example, for non-equality predicates.
10-34
Chapter 10
When the Database Gathers Optimizer Statistics
Note:
The database does not use dynamic statistics for queries that contain the AS
OF clause.
See Also:
See Also:
10-35
11
Histograms
A histogram is a special type of column statistic that provides more detailed
information about the data distribution in a table column. A histogram sorts values into
"buckets," as you might sort coins into buckets.
Based on the NDV and the distribution of the data, the database chooses the type of
histogram to create. (In some cases, when creating a histogram, the database
samples an internally predetermined number of rows.) The types of histograms are as
follows:
• Frequency histograms and top frequency histograms
• Height-Balanced histograms (legacy)
• Hybrid histograms
This chapter contains the following topics:
• Purpose of Histograms
By default the optimizer assumes a uniform distribution of rows across the distinct
values in a column.
• When Oracle Database Creates Histograms
If DBMS_STATS gathers statistics for a table, and if queries have referenced the
columns in this table, then Oracle Database creates histograms automatically as
needed according to the previous query workload.
• How Oracle Database Chooses the Histogram Type
Oracle Database uses several criteria to determine which histogram to create:
frequency, top frequency, height-balanced, or hybrid.
• Cardinality Algorithms When Using Histograms
For histograms, the algorithm for cardinality depends on factors such as the
endpoint numbers and values, and whether column values are popular or
nonpopular.
• Frequency Histograms
In a frequency histogram, each distinct column value corresponds to a single
bucket of the histogram. Because each value has its own dedicated bucket, some
buckets may have many values, whereas others have few.
• Top Frequency Histograms
A top frequency histogram is a variation on a frequency histogram that ignores
nonpopular values that are statistically insignificant.
• Height-Balanced Histograms (Legacy)
In a legacy height-balanced histogram, column values are divided into buckets so
that each bucket contains approximately the same number of rows.
• Hybrid Histograms
A hybrid histogram combines characteristics of both height-based histograms
and frequency histograms. This "best of both worlds" approach enables the
optimizer to obtain better selectivity estimates in some situations.
11-1
Chapter 11
Purpose of Histograms
• As queries change over time, DBMS_STATS may change which statistics it gathers.
For example, even if the data in a table does not change, queries and DBMS_STATS
operations can cause the plans for queries that reference these tables to change.
• If you gather statistics for a table and do not query the table, then the database
does not create histograms for columns in this table. For the database to create
the histograms automatically, you must run one or more queries to populate the
column usage information in SYS.COL_USAGE$.
Example 11-1 Automatic Histogram Creation
Assume that sh.sh_ext is an external table that contains the same rows as the
sh.sales table. You create new table sales2 and perform a bulk load using sh_ext as
11-2
Chapter 11
When Oracle Database Creates Histograms
a source, which automatically creates statistics for sales2. You also create indexes as
follows:
You query the data dictionary to determine whether histograms exist for the sales2
columns. Because sales2 has not yet been queried, the database has not yet created
histograms:
You query sales2 for the number of rows for product 42, and then gather table
statistics using the GATHER AUTO option:
COUNT(*)
----------
12116
A query of the data dictionary now shows that the database created a histogram on
the prod_id column based on the information gather during the preceding query:
11-3
Chapter 11
How Oracle Database Chooses the Histogram Type
Related Topics
• Online Statistics Gathering for Bulk Loads
Starting in Oracle Database 12c, the database can gather table statistics
automatically during the following types of bulk loads: INSERT INTO ... SELECT
into an empty table using a direct path insert, and CREATE TABLE AS SELECT .
The following diagram shows the decision tree for histogram creation.
No No No
11-4
Chapter 11
Cardinality Algorithms When Using Histograms
• Nonpopular values
11-5
Chapter 11
Cardinality Algorithms When Using Histograms
Any value that is not popular is a nonpopular value. The optimizer calculates the
cardinality estimates for nonpopular values using the following formula:
The optimizer calculates density using an internal algorithm based on factors such
as the number of buckets and the NDV. Density is expressed as a decimal number
between 0 and 1. Values close to 1 indicate that the optimizer expects many rows
to be returned by a query referencing this column in its predicate list. Values close
to 0 indicate that the optimizer expects few rows to be returned.
See Also:
Oracle Database Reference to learn about the
DBA_TAB_COL_STATISTICS.DENSITY column
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
1 52792
6 52793
8 52794
9 52795
10 52796
12 52797
14 52798
23 52799
Several buckets are "missing." Originally, buckets 2 through 6 each contained a single
instance of value 52793. The optimizer compressed all of these buckets into the bucket
with the highest endpoint number (bucket 6), which now contains 5 instances of value
52793. This value is popular because the difference between the endpoint number of
the current bucket (6) and the previous bucket (1) is 5. Thus, before compression the
value 52793 was the endpoint for 5 buckets.
The following annotations show which buckets are compressed, and which values are
popular:
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
1 52792 -> nonpopular
6 52793 -> buckets 2-6 compressed into 6; popular
11-6
Chapter 11
Frequency Histograms
11-7
Chapter 11
Frequency Histograms
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
AUTO_SAMPLE_SIZE
Assumptions
This scenario assumes that you want to generate a frequency histogram on the
sh.countries.country_subregion_id column. This table has 23 rows.
The following query shows that the country_subregion_id column contains 8 distinct
values (sample output included) that are unevenly distributed:
COUNTRY_SUBREGION_ID COUNT(*)
-------------------- ----------
52792 1
52793 5
52794 2
52795 1
52796 1
52797 2
52798 2
52799 9
BEGIN
DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'SH'
, tabname => 'COUNTRIES'
, method_opt => 'FOR COLUMNS COUNTRY_SUBREGION_ID'
);
END;
11-8
Chapter 11
Frequency Histograms
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
1 52792
6 52793
8 52794
9 52795
10 52796
12 52797
14 52798
23 52799
Figure 11-2 is a graphical illustration of the 8 buckets in the histogram. Each value
is represented as a coin that is dropped into a bucket.
11-9
Chapter 11
Frequency Histograms
52793 52793
52799
52799 52799
As shown in Figure 11-2, each distinct value has its own bucket. Because this is a
frequency histogram, the endpoint number is the cumulative frequency of
endpoints. For 52793, the endpoint number 6 indicates that the value appears 5
times (6 - 1). For 52794, the endpoint number 8 indicates that the value appears 2
times (8 - 6).
Every bucket whose endpoint is at least 2 greater than the previous endpoint
contains a popular value. Thus, buckets 6, 8, 12, 14, and 23 contain popular
11-10
Chapter 11
Top Frequency Histograms
values. The optimizer calculates their cardinality based on endpoint numbers. For
example, the optimizer calculates the cardinality (C) of value 52799 using the
following formula, where the number of rows in the table is 23:
C = 23 * ( 9 / 23 )
See Also:
11-11
Chapter 11
Top Frequency Histograms
• NDV is greater than n, where n is the number of histogram buckets (default 254).
• The percentage of rows occupied by the top n frequent values is equal to or
greater than threshold p, where p is (1-(1/n))*100.
• The estimate_percent parameter in the DBMS_STATS statistics gathering procedure
is set to AUTO_SAMPLE_SIZE.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
AUTO_SAMPLE_SIZE
Assumptions
This scenario assumes that you want to generate a top frequency histogram on the
sh.countries.country_subregion_id column. This table has 23 rows.
The following query shows that the country_subregion_id column contains 8 distinct
values (sample output included) that are unevenly distributed:
COUNTRY_SUBREGION_ID COUNT(*)
-------------------- ----------
52792 1
52793 5
52794 2
52795 1
52796 1
52797 2
52798 2
52799 9
BEGIN
DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'SH'
, tabname => 'COUNTRIES'
11-12
Chapter 11
Top Frequency Histograms
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
1 52792
6 52793
8 52794
9 52796
11 52797
13 52798
22 52799
11-13
Chapter 11
Top Frequency Histograms
52793 52793
52799
52799 52799
As shown in Figure 11-3, each distinct value has its own bucket except for 52795,
which is excluded from the histogram because it is nonpopular and statistically
insignificant. As in a standard frequency histogram, the endpoint number
represents the cumulative frequency of values.
11-14
Chapter 11
Height-Balanced Histograms (Legacy)
See Also:
11-15
Chapter 11
Height-Balanced Histograms (Legacy)
Assumptions
This scenario assumes that you want to generate a height-balanced histogram on the
sh.countries.country_subregion_id column. This table has 23 rows.
The following query shows that the country_subregion_id column contains 8 distinct
values (sample output included) that are unevenly distributed:
SELECT country_subregion_id, count(*)
FROM sh.countries
GROUP BY country_subregion_id
ORDER BY 1;
COUNTRY_SUBREGION_ID COUNT(*)
-------------------- ----------
52792 1
52793 5
52794 2
52795 1
52796 1
52797 2
52798 2
52799 9
Note:
To simulate Oracle Database 11g behavior, which is necessary to create
a height-based histogram, set estimate_percent to a nondefault value.
If you specify a nondefault percentage, then the database creates
frequency or height-balanced histograms.
11-16
Chapter 11
Height-Balanced Histograms (Legacy)
NUM_OF_ROWS COUNTRY_SUBREGION_ID
----------- --------------------
1 52792
5 52793
2 52794
1 52795
1 52796
2 52797
2 52798
9 52799
4. Query the endpoint number and endpoint value for the country_subregion_id
column.
For example, use the following query (sample output included):
SELECT ENDPOINT_NUMBER, ENDPOINT_VALUE
FROM USER_HISTOGRAMS
WHERE TABLE_NAME='COUNTRIES'
AND COLUMN_NAME='COUNTRY_SUBREGION_ID';
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
0 52792
2 52793
3 52795
4 52798
7 52799
11-17
Chapter 11
Height-Balanced Histograms (Legacy)
52799
52799 52799
The bucket number is identical to the endpoint number. The optimizer records the
value of the last row in each bucket as the endpoint value, and then checks to
ensure that the minimum value is the endpoint value of the first bucket, and the
maximum value is the endpoint value of the last bucket. In this example, the
optimizer adds bucket 0 so that the minimum value 52792 is the endpoint of a
bucket.
The optimizer must evenly distribute 23 rows into the 7 specified histogram
buckets, so each bucket contains approximately 3 rows. However, the optimizer
compresses buckets with the same endpoint. So, instead of bucket 1 containing 2
instances of value 52793, and bucket 2 containing 3 instances of value 52793, the
optimizer puts all 5 instances of value 52793 into bucket 2. Similarly, instead of
having buckets 5, 6, and 7 contain 3 values each, with the endpoint of each bucket
as 52799, the optimizer puts all 9 instances of value 52799 into bucket 7.
In this example, buckets 3 and 4 contain nonpopular values because the
difference between the current endpoint number and previous endpoint number is
1. The optimizer calculates cardinality for these values based on density. The
remaining buckets contain popular values. The optimizer calculates cardinality for
these values based on endpoint numbers.
11-18
Chapter 11
Hybrid Histograms
See Also:
1 1 1 5 5 5 10 10 25 25 25 25 50 100 100
11-19
Chapter 11
Hybrid Histograms
You gather statistics for this table, setting the method_opt argument of
DBMS_STATS.GATHER_TABLE_STATS to FOR ALL COLUMNS SIZE 3. In this case, the
optimizer initially groups the values in the coins column into three buckets, as shown
in the following figure.
1 1 1 5 10 10 25 25 50
5 5 25 25 100 100
If a bucket border splits a value so that some occurrences of the value are in one
bucket and some in another, then the optimizer shifts the bucket border (and all other
following bucket borders) forward to include all occurrences of the value. For example,
the optimizer shifts value 5 so that it is now wholly in the first bucket, and the value 25
is now wholly in the second bucket.
1 1 1 10 10 50
5 5 25 25 100 100
5 25 25
The endpoint repeat count measures the number of times that the corresponding
bucket endpoint, which is the value at the right bucket border, repeats itself. For
example, in the first bucket, the value 5 is repeated 3 times, so the endpoint repeat
count is 3.
11-20
Chapter 11
Hybrid Histograms
1 1 1 10 10 50
5 5 25 25 100 100
5 25 25
11-21
Chapter 11
Hybrid Histograms
See Also:
Assumptions
This scenario assumes that you want to generate a hybrid histogram on the
sh.products.prod_subcategory_id column. This table has 72 rows. The
prod_subcategory_id column contains 22 distinct values.
BEGIN DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'SH'
, tabname => 'PRODUCTS'
, method_opt => 'FOR COLUMNS PROD_SUBCATEGORY_ID SIZE 10'
);
END;
NUM_OF_ROWS PROD_SUBCATEGORY_ID
----------- -------------------
8 2014
7 2055
6 2032
6 2054
5 2056
5 2031
5 2042
5 2051
4 2036
3 2043
11-22
Chapter 11
Hybrid Histograms
2 2033
2 2034
2 2013
2 2012
2 2053
2 2035
1 2022
1 2041
1 2044
1 2011
1 2021
1 2052
22 rows selected.
The column contains 22 distinct values. Because the number of buckets (10) is
less than 22, the optimizer cannot create a frequency histogram. The optimizer
considers both hybrid and top frequency histograms. To qualify for a top frequency
histogram, the percentage of rows occupied by the top 10 most frequent values
must be equal to or greater than threshold p, where p is (1-(1/10))*100, or 90%.
However, in this case the top 10 most frequent values occupy 54 rows out of 72,
which is only 75% of the total. Therefore, the optimizer chooses a hybrid histogram
because the criteria for a top frequency histogram do not apply.
3. Query the histogram information for the country_subregion_id column.
For example, use the following query (sample output included):
4. Query the endpoint number, endpoint value, and endpoint repeat count for the
country_subregion_id column.
For example, use the following query (sample output included):
11-23
Chapter 11
Hybrid Histograms
52 2052 1
54 2053 2
60 2054 6
72 2056 5
10 rows selected.
In a height-based histogram, the optimizer would evenly distribute 72 rows into the
10 specified histogram buckets, so that each bucket contains approximately 7
rows. Because this is a hybrid histogram, the optimizer distributes the values so
that no value occupies more than one bucket. For example, the optimizer does not
put some instances of value 2036 into one bucket and some instances of this value
into another bucket: all instances are in bucket 36.
The endpoint repeat count shows the number of times the highest value in the
bucket is repeated. By using the endpoint number and repeat count for these
values, the optimizer can estimate cardinality. For example, bucket 36 contains
instances of values 2033, 2034, 2035, and 2036. The endpoint value 2036 has an
endpoint repeat count of 4, so the optimizer knows that 4 instances of this value
exist. For values such as 2033, which are not endpoints, the optimizer estimates
cardinality using density.
See Also:
11-24
12
Configuring Options for Optimizer Statistics
Gathering
This chapter explains what optimizer statistics collection is and how to set statistics
preferences.
This chapter contains the following topics:
• About Optimizer Statistics Collection
In Oracle Database, optimizer statistics collection is the gathering of optimizer
statistics for database objects, including fixed objects.
• Setting Optimizer Statistics Preferences
This topic explains how to set optimizer statistics defaults using
DBMS_STATS.SET_*_PREFS procedures.
• Configuring Options for Dynamic Statistics
Dynamic statistics are an optimization technique in which the database uses
recursive SQL to scan a small random sample of the blocks in a table.
• Managing SQL Plan Directives
A SQL plan directive is additional information and instructions that the optimizer
can use to generate a more optimal plan.
12-1
Chapter 12
About Optimizer Statistics Collection
12-2
Chapter 12
Setting Optimizer Statistics Preferences
See Also:
Cloud Control online help
See Also:
Online Help for Oracle Enterprise Manager Cloud Control
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn how to
use DBMS_STATS and DBMS_AUTO_TASK_ADMIN
12-3
Chapter 12
Setting Optimizer Statistics Preferences
You can set optimizer statistics preferences at the table, schema, database (all
tables), and global level. A global preference refers to tables with no preferences and
any tables created in the future. The procedure names follow the form SET_*_PREFS.
12-4
Chapter 12
Setting Optimizer Statistics Preferences
• STALE_PERCENT
This preference determines the percentage of rows in a table that must change
before the database deems the statistics stale and in need of regathering.
• AUTO_STAT_EXTENSIONS
When set to the non-default value of ON, this preference enables a SQL plan
directive to trigger the creation of column group statistics based on usage of
columns in the predicates in the workload.
• INCREMENTAL
This preference determines whether the database maintains the global statistics of
a partitioned table without performing a full table scan. Possible values are TRUE
and FALSE.
For example, by the default setting for INCREMENTAL is FALSE. You can set
INCREMENTAL to TRUE for a range-partitioned table when the last few partitions are
updated. Also, when performing a partition exchange operation on a
nonpartitioned table, Oracle recommends that you set INCREMENTAL to TRUE and
INCREMENTAL_LEVEL to TABLE. With these settings, DBMS_STATS gathers table-level
synopses on this table.
• INCREMENTAL_LEVEL
This preference controls what synopses to collect when INCREMENTAL preference is
set to TRUE. It takes two values: TABLE or PARTITION.
• APPROXIMATE_NDV_ALGORITHM
This preference controls which algorithm to use when calculating the number of
distinct values for partitioned tables using incremental statistics.
Procedure Scope
SET_TABLE_PREFS Specified table only.
12-5
Chapter 12
Setting Optimizer Statistics Preferences
Procedure Scope
SET_SCHEMA_PREFS All existing tables in the specified schema.
This procedure calls SET_TABLE_PREFS for each table in the
specified schema. Calling SET_SCHEMA_PREFS does not affect
any new tables created after it has been run. New tables use the
GLOBAL_PREF values for all parameters.
SET_DATABASE_PREFS All user-defined schemas in the database. You can include
system-owned schemas such as SYS and SYSTEM by setting the
ADD_SYS parameter to true.
This procedure calls SET_TABLE_PREFS for each table in the
specified schema. Calling SET_DATABASE_PREFS does not
affect any new objects created after it has been run. New
objects use the GLOBAL_PREF values for all parameters.
SET_GLOBAL_PREFS Any table that does not have an existing table preference.
All parameters default to the global setting unless a table
preference is set or the parameter is explicitly set in the
DBMS_STATS.GATHER_*_STATS statement. Changes made by
SET_GLOBAL_PREFS affect any new objects created after it runs.
New objects use the SET_GLOBAL_PREFS values for all
parameters.
With SET_GLOBAL_PREFS, you can set a default value for the
parameter AUTOSTATS_TARGET. This additional parameter
controls which objects the automatic statistic gathering job
running in the nightly maintenance window affects. Possible
values for AUTOSTATS_TARGET are ALL, ORACLE, and AUTO
(default).
You can only set the CONCURRENT preference at the global level.
You cannot set the preference INCREMENTAL_LEVEL using
SET_GLOBAL_PREFS.
See Also:
12-6
Chapter 12
Setting Optimizer Statistics Preferences
statistics preference. In this way, you control when the database honors a parameter
value passed to the statistics gathering procedures.
When preference_overrides_parameter is set to FALSE (default), the input values for
statistics gathering procedures are honored. When set to TRUE, the input values are
ignored.
Set the preference_overrides_parameter preference using the SET_TABLE_PREFS,
SET_SCHEMA_PREFS, or SET_GLOBAL_PREFS procedures in DBMS_STATS. Regardless of
whether preference_overrides_parameter is set, the database uses the same order
of precedence for setting statistics:
1. Table preference (set for a specific table, all tables in a schema, or all tables in the
database)
2. Global preference
3. Default preference
Example 12-1 Overriding Statistics Preferences at the Table Level
In this example, legacy scripts set estimate_percent explicitly rather than using the
recommended AUTO_SAMPLE_SIZE. Your goal is to prevent users from using these
scripts to set preferences on the sh.costs table.
Action Description
STAT_PREFS
----------
DBMS_STATS.AUTO_SAMPLE_SIZE
12-7
Chapter 12
Setting Optimizer Statistics Preferences
Action Description
STAT_PREFS
----------
DBMS_STATS.AUTO_SAMPLE_SIZE
12-8
Chapter 12
Setting Optimizer Statistics Preferences
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS procedures for setting optimizer statistics
Action Description
SQL> SELECT DBMS_STATS.GET_PREFS You query the INCREMENTAL preference for costs and
('incremental', 'sh','costs') determine that it is set to true.
AS "STAT_PREFS" FROM DUAL;
STAT_PREFS
----------
TRUE
SQL> SELECT DBMS_STATS.GET_PREFS You query the INCREMENTAL preference for costs and
('incremental', 'sh', 'costs') confirm that it is set to false.
AS "STAT_PREFS" FROM DUAL;
STAT_PREFS
----------
FALSE
SQL> SELECT DBMS_STATS.GET_PREFS You query the INCREMENTAL preference for costs and
('incremental', 'sh', 'costs') confirm that it is set to true.
AS "STAT_PREFS" FROM DUAL;
STAT_PREFS
----------
TRUE
12-9
Chapter 12
Setting Optimizer Statistics Preferences
Action Description
SQL> SELECT DBMS_STATS.GET_PREFS You query the INCREMENTAL preference for costs and
('incremental', 'sh', 'costs') confirm that it is set to false.
AS "STAT_PREFS" FROM DUAL;
STAT_PREFS
----------
FALSE
See Also:
Online Help for Oracle Enterprise Manager Cloud Control
12-10
Chapter 12
Setting Optimizer Statistics Preferences
See Also:
Online Help for Oracle Enterprise Manager Cloud Control
Prerequisites
This task has the following prerequisites:
• To set the global or database preferences, you must have SYSDBA privileges, or
both ANALYZE ANY DICTIONARY and ANALYZE ANY system privileges.
12-11
Chapter 12
Setting Optimizer Statistics Preferences
3. Execute the appropriate procedure from Table 12-1, specifying the following
parameters:
• ownname - Set schema name (SET_TAB_PREFS and SET_SCHEMA_PREFS only)
• tabname - Set table name (SET_TAB_PREFS only)
• pname - Set parameter name
• pvalue - Set parameter value
• add_sys - Include system tables (optional, SET_DATABASE_PREFS only)
The following example specifies that 13% of rows in sh.sales must change before
the statistics on that table are considered stale:
12-12
Chapter 12
Configuring Options for Dynamic Statistics
See Also:
Oracle Database PL/SQL Packages and Types Reference for descriptions of
the parameter names and values for program units
Note:
Dynamic statistics were called dynamic sampling in releases earlier than
Oracle Database 12c Release 1 (12.1).
12-13
Chapter 12
Configuring Options for Dynamic Statistics
The following table describes the levels for dynamic statistics. Note the following:
• If OPTIMIZER_DYNAMIC_STATISTICS is TRUE, and if dynamic statistics are not
disabled, then the database may choose to use dynamic statistics when a SQL
statement uses parallel execution.
• If OPTIMIZER_ADAPTIVE_STATISTICS is TRUE, then the optimizer uses dynamic
statistics when relevant SQL plan directives exist. The database maintains the
resulting statistics in the SQL plan directives store, making them available to other
queries.
Level When the Optimizer Uses Dynamic Statistics Sample Size (Blocks)
0 Do not use dynamic statistics. n/a
1 Use dynamic statistics for all tables that do not have statistics, but 32
only if the following criteria are met:
• At least one nonpartitioned table in the query does not have
statistics.
• This table has no indexes.
• This table has more blocks than the number of blocks that
would be used for dynamic statistics of this table.
2 Use dynamic statistics if at least one table in the statement has no 64
statistics. This is the default value.
3 Use dynamic statistics if any of the following conditions is true: 64
• At least one table in the statement has no statistics.
• The statement has one or more expressions used in the WHERE
clause predicates, for example, WHERE
SUBSTR(CUSTLASTNAME,1,3).
4 Use dynamic statistics if any of the following conditions is true: 64
• At least one table in the statement has no statistics.
• The statement has one or more expressions used in the WHERE
clause predicates, for example, WHERE
SUBSTR(CUSTLASTNAME,1,3).
• The statement uses complex predicates (an OR or AND
operator between multiple predicates on the same table).
5 The criteria are identical to level 4, but the database uses a 128
different sample size.
6 The criteria are identical to level 4, but the database uses a 256
different sample size.
7 The criteria are identical to level 4, but the database uses a 512
different sample size.
8 The criteria are identical to level 4, but the database uses a 1024
different sample size.
9 The criteria are identical to level 4, but the database uses a 4086
different sample size.
10 The criteria are identical to level 4, but the database uses a All blocks
different sample size.
11 The database uses adaptive dynamic sampling automatically when Automatically determined
the optimizer deems it necessary.
12-14
Chapter 12
Configuring Options for Dynamic Statistics
See Also:
Assumptions
This tutorial assumes the following:
• You want correct selectivity estimates for the following query, which has WHERE
clause predicates on two correlated columns:
SELECT *
FROM sh.customers
WHERE cust_city='Los Angeles'
AND cust_state_province='CA';
12-15
Chapter 12
Configuring Options for Dynamic Statistics
SELECT *
FROM TABLE(DBMS_XPLAN.DISPLAY);
The output appears below (the example has been reformatted to fit on the page):
-------------------------------------------------------------------------------
|Id| Operation | Name |Rows|Bytes|Cost | Time |
-------------------------------------------------------------------------------
| 0| SELECT STATEMENT | | 53| 9593|53(0)|00:00:01|
| 1| TABLE ACCESS BY INDEX ROWID|CUSTOMERS | 53| 9593|53(0)|00:00:01|
|*2| INDEX RANGE SCAN |CUST_CITY_STATE_IND| 53| 9593| 3(0)|00:00:01|
-------------------------------------------------------------------------------
The columns in the WHERE clause have a real-world correlation, but the optimizer is
not aware that Los Angeles is in California and assumes both predicates reduce
the number of rows returned. Thus, the table contains 932 rows that meet the
conditions, but the optimizer estimates 53, as shown in bold.
If the database had used dynamic statistics for this plan, then the Note section of
the plan output would have indicated this fact. The optimizer did not use dynamic
statistics because the statement executed serially, standard statistics exist, and
the parameter OPTIMIZER_DYNAMIC_SAMPLING is set to the default of 2.
4. Set the dynamic statistics level to 4 in the session using the following statement:
The new plan shows a more accurate estimate of the number of rows, as shown
by the value 932 in bold:
PLAN_TABLE_OUTPUT
------------------------------------------------------------------------
---
Plan hash value: 2008213504
------------------------------------------------------------------------
---
| Id | Operation | Name |Rows | Bytes |Cost (%CPU)|
Time |
------------------------------------------------------------------------
---
12-16
Chapter 12
Configuring Options for Dynamic Statistics
Note
-----
- dynamic statistics used for this statement (level=4)
The note at the bottom of the plan indicates that the sampling level is 4. The
additional dynamic statistics made the optimizer aware of the real-world
relationship between the cust_city and cust_state_province columns, thereby
enabling it to produce a more accurate estimate for the number of rows: 932 rather
than 53.
See Also:
12-17
Chapter 12
Managing SQL Plan Directives
See Also:
Oracle Database Reference to learn about the
OPTIMIZER_DYNAMIC_SAMPLING initialization parameter
Procedure Description
FLUSH_SQL_PLAN_DIRECTIVE Forces the database to write directives from memory to
persistent storage in the SYSAUX tablespace.
DROP_SQL_PLAN_DIRECTIVE Drops a SQL plan directive. If a directive that triggers
dynamic sampling is creating unacceptable performance
overhead, then you may want to remove it manually.
If a SQL plan directive is dropped manually or
automatically, then the database can re-create it. To
prevent its re-creation, you can use
DBMS_SPM.ALTER_SQL_PLAN_DIRECTIVE to do the
following:
• Disable the directive by setting ENABLED to NO
• Prevent the directive from being dropped by setting
AUTO_DROP to NO
To disable SQL plan directives, set
OPTIMIZER_ADAPTIVE_STATISTICS to FALSE.
Prerequisites
You must have the Administer SQL Management Object privilege to execute the
DBMS_SPD APIs.
Assumptions
This tutorial assumes that you want to do the following:
• Write all directives for the sh schema to persistent storage.
• Delete all directives for the sh schema.
12-18
Chapter 12
Managing SQL Plan Directives
BEGIN
DBMS_SPD.FLUSH_SQL_PLAN_DIRECTIVE;
END;
/
3. Query the data dictionary for information about existing directives in the sh
schema.
Example 12-3 queries the data dictionary for information about the directive.
4. Delete the existing SQL plan directive for the sh schema.
The following PL/SQL program unit deletes the SQL plan directive with the ID
1484026771529551585:
BEGIN
DBMS_SPD.DROP_SQL_PLAN_DIRECTIVE ( directive_id =>
1484026771529551585 );
END;
/
12-19
Chapter 12
Managing SQL Plan Directives
See Also:
12-20
13
Gathering Optimizer Statistics
This chapter explains how to use the DBMS_STATS.GATHER_*_STATS program units.
See Also:
13-1
Chapter 13
Configuring Automatic Optimizer Statistics Collection
Note:
Data visibility and privilege requirements may differ when using automatic
optimizer statistics collection with pluggable databases.
To collect the optimizer statistics, the database calls an internal procedure that
operates similarly to the GATHER_DATABASE_STATS procedure with the GATHER AUTO
option. Automatic statistics collection honors all preferences set in the database.
The principal difference between manual and automatic collection is that the latter
prioritizes database objects that need statistics. Before the maintenance window
closes, automatic collection assesses all objects and prioritizes objects that have no
statistics or very old statistics.
Note:
When gathering statistics manually, you can reproduce the object
prioritization of automatic collection by using the DBMS_AUTO_TASK_IMMEDIATE
package. This package runs the same statistics gathering job that is
executed during the automatic nightly statistics gathering job.
See Also:
Oracle Database Administrator’s Guide for a table that summarizes how
manageability features work in a container database (CDB)
13-2
Chapter 13
Configuring Automatic Optimizer Statistics Collection
potential conflicts that result from operations occurring at the same time as automatic
statistics collection, Oracle recommends that you change the window accordingly.
Prerequisites
Access the Database Home page, as described in "Accessing the Database Home
Page in Cloud Control."
Note:
Oracle strongly recommends that you not disable automatic statistics
gathering because it is critical for the optimizer to generate optimal
plans for queries against dictionary and user objects. If you disable
automatic collection, ensure that you have a good manual statistics
collection strategy for dictionary and user schemas.
b. To disable statistics gathering for specific days in the week, check the
appropriate box next to the window name.
c. To change the characteristics of a window group, click Edit Window Group.
d. To change the times for a window, click the name of the window (for example,
MONDAY_WINDOW), and then in the Schedule section, click Edit.
The Edit Window page appears.
13-3
Chapter 13
Configuring Automatic Optimizer Statistics Collection
In this page, you can change the parameters such as duration and start time
for window execution.
e. Click Apply.
See Also:
Online Help for Oracle Enterprise Manager Cloud Control
13-4
Chapter 13
Configuring Automatic Optimizer Statistics Collection
Note:
Because monitoring and many automatic features are disabled, Oracle
strongly recommends that you do not set STATISTICS_LEVEL to BASIC.
BEGIN
DBMS_AUTO_TASK_ADMIN.ENABLE (
client_name => 'auto optimizer stats collection'
, operation => NULL
, window_name => NULL
);
END;
/
BEGIN
DBMS_AUTO_TASK_ADMIN.DISABLE (
client_name => 'auto optimizer stats collection'
, operation => NULL
, window_name => NULL
);
END;
/
CLIENT_NAME STATUS
------------------------------- --------
auto optimizer stats collection ENABLED
13-5
Chapter 13
Gathering Optimizer Statistics Manually
BEGIN
DBMS_SCHEDULER.SET_ATTRIBUTE (
'MONDAY_WINDOW'
, 'repeat_interval'
, 'freq=daily;byday=MON;byhour=05;byminute=0;bysecond=0'
);
END;
/
See Also:
13-6
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
Procedure Purpose
GATHER_INDEX_STATS Collects index statistics
GATHER_TABLE_STATS Collects table, column, and index statistics
GATHER_SCHEMA_STATS Collects statistics for all objects in a schema
GATHER_DICTIONARY_STATS Collects statistics for all system schemas, including SYS
and SYSTEM, and other optional schemas, such as
CTXSYS and DRSYS
GATHER_DATABASE_STATS Collects statistics for all objects in a database
When the OPTIONS parameter is set to GATHER STALE or GATHER AUTO, the
GATHER_SCHEMA_STATS and GATHER_DATABASE_STATS procedures gather statistics for
any table that has stale statistics and any table that is missing statistics. If a monitored
table has been modified more than 10%, then the database considers these statistics
stale and gathers them again.
13-7
Chapter 13
Gathering Optimizer Statistics Manually
Note:
As explained in "Configuring Automatic Optimizer Statistics Collection", you
can configure a nightly job to gather statistics automatically.
See Also:
This section offers guidelines for typical situations in which you may choose to gather
statistically manually:
• Guideline for Setting the Sample Size
In the context of optimizer statistics, sampling is the gathering of statistics from a
random subset of table rows. By enabling the database to avoid full table scans
and sorts of entire tables, sampling minimizes the resources necessary to gather
statistics.
• Guideline for Gathering Statistics in Parallel
By default, the database gathers statistics with the parallelism degree specified at
the table or index level.
13-8
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
• "Hybrid Histograms"
• Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.AUTO_SAMPLE_SIZE
13-9
Chapter 13
Gathering Optimizer Statistics Manually
Note:
Do not confuse gathering statistics in parallel with gathering statistics
concurrently.
See Also:
See Also:
"Gathering Incremental Statistics on Partitioned Objects"
13-10
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
13-11
Chapter 13
Gathering Optimizer Statistics Manually
Note:
Starting in Oracle Database 12c Release 2 (12.2), you no longer need to use
DBMS_STATS.FLUSH_DATABASE_MONITORING_INFO to ensure that view
metadata is current. The statistics shown in the DBA_TAB_STATISTICS,
DBA_IND_STATISTICS, and DBA_TAB_MODIFICATIONS views are obtained from
both disk and memory.
• YES
The statistics are stale.
• NO
The statistics are not stale.
• null
The statistics are not collected.
Executing GATHER_SCHEMA_STATS or GATHER_DATABASE_STATS with the GATHER AUTO
option collects statistics only for objects with no statistics or stale statistics.
PARTITION_NAME STA
--------------- ---
SALES_1995 NO
SALES_1996 NO
SALES_H1_1997 NO
SALES_H2_1997 NO
SALES_Q1_1998 NO
SALES_Q1_1999 NO
.
.
.
13-12
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
Oracle Database Reference to learn about the DBA_TAB_MODIFICATIONS view
BEGIN
DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'sh'
, tabname => 'customers'
, degree => 2
);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
GATHER_TABLE_STATS procedure
13-13
Chapter 13
Gathering Optimizer Statistics Manually
missing. Instead, the optimizer uses predefined default values. These defaults may not
be representative and could potentially lead to a suboptimal execution plan. Thus, it is
important to keep fixed object statistics current.
Oracle Database automatically gathers fixed object statistics as part of automated
statistics gathering if they have not been previously collected. You can also manually
collect statistics on fixed objects by calling DBMS_STATS.GATHER_FIXED_OBJECTS_STATS.
Oracle recommends that you gather statistics when the database has representative
activity.
Prerequisites
You must have the SYSDBA or ANALYZE ANY DICTIONARY system privilege to execute
this procedure.
BEGIN
DBMS_STATS.GATHER_FIXED_OBJECTS_STATS;
END;
/
See Also:
13-14
Chapter 13
Gathering Optimizer Statistics Manually
Assumptions
This tutorial assumes the following:
• The oe.orders table is extremely volatile.
• You want to delete and then lock the statistics on the orders table to prevent the
database from gathering statistics on the table. In this way, the database can
dynamically gather necessary statistics as part of query optimization.
• The oe user has the necessary privileges to query DBMS_XPLAN.DISPLAY_CURSOR.
BEGIN
DBMS_STATS.DELETE_TABLE_STATS('OE','ORDERS');
END;
/
BEGIN
DBMS_STATS.LOCK_TABLE_STATS('OE','ORDERS');
END;
/
---------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
---------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | 2 (100)| |
13-15
Chapter 13
Gathering Optimizer Statistics Manually
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| ORDERS | 105 | 2 (0)| 00:00:01 |
---------------------------------------------------------------------
Note
-----
- dynamic statistics used for this statement (level=2)
The Note in the preceding execution plan shows that the database used dynamic
statistics for the SELECT statement.
See Also:
Note:
Concurrent statistics gathering mode does not rely on parallel query
processing, but is usable with it.
13-16
Chapter 13
Gathering Optimizer Statistics Manually
The database runs as many concurrent jobs as possible. The Job Scheduler decides
how many jobs to execute concurrently and how many to queue. As running jobs
complete, the scheduler dequeues and runs more jobs until the database has
gathered statistics on all tables, partitions, and subpartitions. The maximum number of
jobs is bounded by the JOB_QUEUE_PROCESSES initialization parameter and available
system resources.
In most cases, the DBMS_STATS procedures create a separate job for each table
partition or subpartition. However, if the partition or subpartition is empty or very small,
then the database may automatically batch the object with other small objects into a
single job to reduce the overhead of job maintenance.
The following figure illustrates the creation of jobs at different levels, where Table 3 is
a partitioned table, and the other tables are nonpartitioned. Job 3 acts as a coordinator
job for Table 3, and creates a job for each partition in that table, and a separate job for
the global statistics of Table 3. This example assumes that incremental statistics
gathering is disabled; if enabled, then the database derives global statistics from
partition-level statistics after jobs for partitions complete.
13-17
Chapter 13
Gathering Optimizer Statistics Manually
Level 1
Level 2
See Also:
13-18
Chapter 13
Gathering Optimizer Statistics Manually
Note:
The ORA$AUTOTASK consumer group is shared with the maintenance tasks
that automatically run during the maintenance windows. Thus, when
concurrency is activated for automatic statistics gathering, the database
automatically manages resources, with no extra steps required.
See Also:
Oracle Database Administrator’s Guide to learn about the Resource
Manager
Prerequisites
This tutorial has the following prerequisites:
• In addition to the standard privileges for gathering statistics, you must have the
following privileges:
– CREATE JOB
– MANAGE SCHEDULER
– MANAGE ANY QUEUE
• The SYSAUX tablespace must be online because the scheduler stores its internal
tables and views in this tablespace.
• The JOB_QUEUE_PROCESSES initialization parameter must be set to at least 4.
• The Resource Manager must be enabled.
13-19
Chapter 13
Gathering Optimizer Statistics Manually
By default, the Resource Manager is disabled. If you do not have a resource plan,
then consider enabling the Resource Manager with the system-supplied
DEFAULT_PLAN.
Assumptions
This tutorial assumes that you want to do the following:
• Enable concurrent statistics gathering
• Gather statistics for the sh schema
• Monitor the gathering of the sh statistics
BEGIN
DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','ALL');
13-20
Chapter 13
Gathering Optimizer Statistics Manually
END;
/
DBMS_STATS.GET_PREFS('CONCURRENT')
----------------------------------
ALL
EXEC DBMS_STATS.GATHER_SCHEMA_STATS('SH');
EXEC DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','OFF');
13-21
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
All statistics gathering job names start with the string ST$.
13-22
Chapter 13
Gathering Optimizer Statistics Manually
VARIABLE id NUMBER
EXEC :id := 981
See Also:
Oracle Database Reference to learn about the DBA_SCHEDULER_JOBS view
13-23
Chapter 13
Gathering Optimizer Statistics Manually
13-24
Chapter 13
Gathering Optimizer Statistics Manually
database avoids the two full scans that are required when not using incremental
statistics: one scan for the partition-level statistics, and one scan for the global-
level statistics.
• In subsequent statistics gathering, the database only needs to scan the stale
partitions and update their statistics (including synopses). The database can
derive global statistics from the fresh partition statistics, which saves a full table
scan.
When using incremental statistics, the database must still gather statistics on any
partition that will change the global or table-level statistics. Incremental statistics
maintenance yields the same statistics as gathering table statistics from scratch, but
performs better.
Calculating the NDV in the table by adding the NDV of the individual partitions
produces an NDV of 9, which is incorrect. Thus, a more accurate technique is
required: synopses.
This section contains the following topics:
• Partition-Level Synopses
A synopsis is special type of statistic that tracks the number of distinct values
(NDV) for each column in a partition. You can consider a synopsis as an internal
management structure that samples distinct values.
• NDV Algorithms: Adaptive Sampling and HyperLogLog
Starting in Oracle Database 12c Release 2 (12.2), the HyperLogLog algorithm can
improve NDV (number of distinct values) calculation performance, and also reduce
the storage space required for synopses.
• Aggregation of Global Statistics Using Synopses: Example
In this example, the database gathers statistics for the initial six partitions of the
sales table, and then creates synopses for each partition (S1, S2, and so on). The
database creates global statistics by aggregating the partition-level statistics and
synopses.
13-25
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
Oracle Database Reference to learn more about DBA_PART_COL_STATISTICS
NDV = B * 2^S
Adaptive sampling produces accurate NDV statistics, but has the following
consequences:
• Synopses occupy significant disk space, especially when tables have many
columns and partitions, and the NDV in each column is high.
13-26
Chapter 13
Gathering Optimizer Statistics Manually
For example, a 60-column table might have 300,000 partitions, with an average
per-column NDV of 5,000. In this example, each partition has 300,000 entries (60
x 5000). In total, the synopses tables have 90 billion entries (300,000 squared),
which occupies at least 720 GB of storage space.
• Bulk processing of synopses can negatively affect performance.
Before the database regathers statistics on the stale partitions, it must delete the
associated synopses. Bulk deletion can be slow because it generates significant
amounts of undo and redo data.
In contrast to dynamic sampling, the HyperLogLog algorithm uses a randomization
technique. Although the algorithm is complex, the foundational insight is that in a
stream of random values, n distinct values will be spaced on average 1/n apart.
Therefore, if you know the smallest value in the stream, then you can roughly estimate
the number of distinct values. For example, if the values range from 0 to 1, and if the
smallest value you observe is .2, then the numbers will on average be evenly spaced .
2 apart, so the NDV estimate is 5.
The HyperLogLog algorithm expands on and corrects the original estimate. The
database applies a hash function to every column value, resulting in a set of hash
values with the same cardinality as the column. For the base estimate, the NDV
equals 2n, where n is the maximum number of trailing zeroes observed in the binary
representation of the hash values. The database refines its NDV estimate by using
part of the output to split values into different hash buckets.
The advantages of the HyperLogLog algorithm over adaptive sampling are:
• The accuracy of the new algorithm is similar to the original algorithm.
• The memory required is significantly lower, which typically leads to huge
reductions in synopsis size.
Synopses can become large when many partitions exist, and they have many
columns with high NDV. Synopses that use the HyperLogLog algorithm are more
compact. Creating and deleting synopses affects batch run times. Any operational
procedures that manage partitions reduce run time.
The DBMS_STATS preference APPROXIMATE_NDV_ALGORITHM determines which algorithm
the database uses for NDV calculation.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
APPROXIMATE_NDV_ALGORITHM preference
13-27
Chapter 13
Gathering Optimizer Statistics Manually
May 18 2012 S1
2 The database generates
1
global statistics by
May 19 2012 S2 aggregating partition-level
statistics and synopses
May 20 2012 S3
Global
Statistics
May 21 2012 S4
May 22 2012 S5
May 23 2012 S6
Sysaux
Tablespace
The following graphic shows a new partition, containing data for May 24, being added
to the sales table. The database gathers statistics for the newly added partition,
retrieves synopses for the other partitions, and then aggregates the synopses to
create global statistics.
Sales Table
May 18 2012 S1
1 The database generates
6
global statistics by
May 19 2012 S2 aggregating partition-level
synopses.
May 20 2012 S3
Global
Statistics
May 21 2012 S4
May 24 2012 S7
Sysaux
Tablespace
13-28
Chapter 13
Gathering Optimizer Statistics Manually
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS
13-29
Chapter 13
Gathering Optimizer Statistics Manually
INCREMENTAL is set to false, which is the default, the database uses a full table scan to
maintain global statistics.
For the database to update global statistics incrementally by scanning only the
partitions that have changed, the following conditions must be met:
• The PUBLISH value for the partitioned table is true.
• The INCREMENTAL value for the partitioned table is true.
• The statistics gathering procedure must specify AUTO_SAMPLE_SIZE for
ESTIMATE_PERCENT and AUTO for GRANULARITY.
Example 13-3 Enabling Incremental Statistics
Assume that the PUBLISH value for the partitioned table sh.sales is true. The
following program enables incremental statistics for this table:
• ALLOW_MIXED_FORMAT
This is the default. If this value is specified, and if the following conditions are met,
then the database does not consider existing adaptive sampling synopses as
stale:
– The synopses are fresh.
– You gather statistics manually.
Thus, synopses in both the legacy and HyperLogLog formats can co-exist.
However, over time the automatic statistics gathering job regathers statistics on
synopses that use the old format, and replaces them with synopses in
13-30
Chapter 13
Gathering Optimizer Statistics Manually
HyperLogLog format. In this way, the automatic statistics gather job gradually
phases out the old format. Manual statistics gathering jobs do not reformat
synopses that use the adaptive sampling format.
• Null
Any partitions with the synopses in the legacy format are considered stale, which
immediately triggers the database to regather statistics for stale synopses. The
advantage is that the performance cost occurs only once. The disadvantage is that
regathering all statistics on large tables can be resource-intensive.
BEGIN
DBMS_STATS.SET_TABLE_PREFS
( ownname => 'sh'
, tabname => 'sales'
, pname => 'approximate_ndv_algorithm'
, pvalue => 'hyperloglog' );
DBMS_STATS.SET_TABLE_PREFS
( ownname => 'sh'
, tabname => 'sales'
, pname => 'incremental_staleness'
, pvalue => 'allow_mixed_format' );
END;
BEGIN
DBMS_STATS.SET_TABLE_PREFS
( ownname => 'sh'
, tabname => 'sales'
, pname => 'approximate_ndv_algorithm'
13-31
Chapter 13
Gathering Optimizer Statistics Manually
DBMS_STATS.SET_TABLE_PREFS
( ownname => 'sh'
, tabname => 'sales'
, pname => 'incremental_staleness'
, pvalue => 'null' );
END;
Assumptions
This tutorial assumes the following:
• You want to load empty partition p_sales_01_2010 in a sales table.
• You create a staging table t_sales_01_2010, and then populate the table.
• You want the database to maintain incremental statistics as part of the partition
exchange operation without having to explicitly gather statistics on the partition
after the exchange.
13-32
Chapter 13
Gathering Optimizer Statistics Manually
BEGIN
DBMS_STATS.SET_TABLE_PREFS (
ownname => 'sh'
, tabname => 't_sales_01_2010'
, pname => 'INCREMENTAL'
, pvalue => 'true'
);
DBMS_STATS.SET_TABLE_PREFS (
ownname => 'sh'
, tabname => 't_sales_01_2010'
, pname => 'INCREMENTAL_LEVEL'
, pvalue => 'table'
);
END;
BEGIN
DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'SH'
, tabname => 'T_SALES_01_2010'
);
END;
/
BEGIN
DBMS_STATS.SET_TABLE_PREFS (
ownname => 'sh'
, tabname => 'sales'
, pname => 'INCREMENTAL'
, pvalue => 'true'
);
END;
/
4. If you have never gathered statistics on sh.sales before with INCREMENTAL set to
true, then gather statistics on the partition to be exchanged.
For example, run the following PL/SQL code:
BEGIN
DBMS_STATS.GATHER_TABLE_STATS (
ownname => 'sh'
, tabname => 'sales'
, pname => 'p_sales_01_2010'
, pvalue => granularity=>'partition'
13-33
Chapter 13
Gathering Optimizer Statistics Manually
);
END;
/
After the exchange, the partitioned table has both statistics and a synopsis for
partition p_sales_01_2010.
In releases before Oracle Database 12c, the preceding statement swapped the
segment data and statistics of p_sales_01_2010 with t_sales_01_2010. The
database did not maintain synopses for nonpartitioned tables such as
t_sales_01_2010. To gather global statistics on the partitioned table, you needed
to rescan the p_sales_01_2010 partition to obtain its synopses.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.SET_TABLE_PREFS
13-34
Chapter 13
Gathering Optimizer Statistics Manually
BEGIN
DBMS_STATS.SET_TABLE_PREFS (
ownname => null
, table_name => 't'
, pname => 'incremental_staleness'
, pvalue => 'use_stale_percent,use_locked_stats'
);
END;
Assumptions
This tutorial assumes the following:
• The STALE_PERCENT for a partitioned table is set to 10.
• The INCREMENTAL value is set to true.
• The table has had statistics gathered in INCREMENTAL mode before.
• You want to discover how statistics gathering changes depending on the setting
for INCREMENTAL_STALENESS, whether the statistics are locked, and the percentage
of DML changes.
13-35
Chapter 13
Gathering System Statistics Manually
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.SET_TABLE_PREFS
13-36
Chapter 13
Gathering System Statistics Manually
When the database gathers system statistics, it analyzes activity in a specified time
period (workload statistics) or simulates a workload (noworkload statistics). The input
arguments to DBMS_STATS.GATHER_SYSTEM_STATS are:
• NOWORKLOAD
The optimizer gathers statistics based on system characteristics only, without
regard to the workload.
• INTERVAL
After the specified number of minutes has passed, the optimizer updates system
statistics either in the data dictionary, or in an alternative table (specified by
stattab). Statistics are based on system activity during the specified interval.
• START and STOP
START initiates gathering statistics. STOP calculates statistics for the elapsed period
(since START) and refreshes the data dictionary or an alternative table (specified by
stattab). The optimizer ignores INTERVAL.
• EXADATA
The system statistics consider the unique capabilities provided by using Exadata,
such as large I/O size and high I/O throughput. The optimizer sets the multiblock
read count and I/O throughput statistics along with CPU speed.
The following table lists the optimizer system statistics gathered by DBMS_STATS and
the options for gathering or manually setting specific system statistics.
13-37
Chapter 13
Gathering System Statistics Manually
13-38
Chapter 13
Gathering System Statistics Manually
See Also:
Oracle Database PL/SQL Packages and Types Reference for detailed
information on the procedures in the DBMS_STATS package for implementing
system statistics
13-39
Chapter 13
Gathering System Statistics Manually
When gathering workload statistics, the database may not gather the mbrc and
mreadtim values if no table scans occur during serial workloads, as is typical of OLTP
systems. However, full table scans occur frequently on DSS systems. These scans
may run parallel and bypass the buffer cache. In such cases, the database still gathers
the sreadtim because index lookups use the buffer cache.
If the database cannot gather or validate gathered mbrc or mreadtim values, but has
gathered sreadtim and cpuspeed, then the database uses only sreadtim and
cpuspeed for costing. In this case, the optimizer uses the value of the initialization
parameter DB_FILE_MULTIBLOCK_READ_COUNT to cost a full table scan. However, if
DB_FILE_MULTIBLOCK_READ_COUNT is 0 or is not set, then the optimizer uses a value of
8 for calculating cost.
13-40
Chapter 13
Gathering System Statistics Manually
See Also:
Oracle Database Reference to learn about the
DB_FILE_MULTIBLOCK_READ_COUNT initialization parameter
Assumptions
This tutorial assumes the following:
• The hour between 10 a.m. and 11 a.m. is representative of the daily workload.
• You intend to collect system statistics directly in the data dictionary.
The optimizer can now use the workload statistics to generate execution plans that
are effective during the normal daily workload.
5. Optionally, query the system statistics.
For example, run the following query:
13-41
Chapter 13
Gathering System Statistics Manually
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GATHER_SYSTEM_STATS procedure
Assumptions
This tutorial assumes the following:
• The database application processes OLTP transactions during the day and runs
OLAP reports at night. To gather representative statistics, you collect them during
the day for two hours and then at night for two hours.
• You want to store statistics in a table named workload_stats.
• You intend to switch between the statistics gathered.
BEGIN
DBMS_STATS.CREATE_STAT_TABLE (
ownname => 'dba1'
, stattab => 'workload_stats'
);
END;
/
BEGIN
DBMS_STATS.GATHER_SYSTEM_STATS (
interval => 120
, stattab => 'workload_stats'
, statid => 'OLTP'
);
13-42
Chapter 13
Gathering System Statistics Manually
END;
/
BEGIN
DBMS_STATS.GATHER_SYSTEM_STATS (
interval => 120
, stattab => 'workload_stats'
, statid => 'OLAP'
);
END;
/
6. In the day or evening, import the appropriate statistics into the data dictionary.
For example, during the day you can import the OLTP statistics from the staging
table into the dictionary with the following program:
BEGIN
DBMS_STATS.IMPORT_SYSTEM_STATS (
stattab => 'workload_stats'
, statid => 'OLTP'
);
END;
/
For example, during the night you can import the OLAP statistics from the staging
table into the dictionary with the following program:
BEGIN
DBMS_STATS.IMPORT_SYSTEM_STATS (
stattab => 'workload_stats'
, statid => 'OLAP'
);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GATHER_SYSTEM_STATS procedure
13-43
Chapter 13
Gathering System Statistics Manually
Assumptions
This tutorial assumes that you want to gather noworkload statistics manually.
BEGIN
DBMS_STATS.GATHER_SYSTEM_STATS (
gathering_mode => 'NOWORKLOAD'
);
END;
13-44
Chapter 13
Running Statistics Gathering Functions in Reporting Mode
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GATHER_SYSTEM_STATS procedure
This procedure deletes workload statistics collected using the INTERVAL or START and
STOP options, and then resets the default to noworkload statistics. However, if the
stattab parameter specifies a table for storing statistics, then the subprogram deletes
all system statistics with the associated statid from the statistics table.
Assumptions
This tutorial assumes the following:
• You gathered statistics for a specific intensive workload, but no longer want the
optimizer to use these statistics.
• You stored workload statistics in the default location, not in a user-specified table.
EXEC DBMS_STATS.DELETE_SYSTEM_STATS;
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.DELETE_SYSTEM_STATS procedure
When you use the REPORT_* procedures, the optimizer does not actually gather
statistics. Rather, the package reports objects that would be processed if you were to
use a specified statistics gathering function.
The following table lists the DBMS_STATS.REPORT_GATHER_*_STATS functions. For all
functions, the input parameters are the same as for the corresponding
13-45
Chapter 13
Running Statistics Gathering Functions in Reporting Mode
Function Description
REPORT_GATHER_TABLE_STATS Runs GATHER_TABLE_STATS in reporting mode.
The procedure does not collect statistics, but
reports all objects that would be affected by
invoking GATHER_TABLE_STATS.
REPORT_GATHER_SCHEMA_STATS Runs GATHER_SCHEMA_STATS in reporting mode.
The procedure does not actually collect statistics,
but reports all objects that would be affected by
invoking GATHER_SCHEMA_STATS.
REPORT_GATHER_DICTIONARY_STATS Runs GATHER_DICTIONARY_STATS in reporting
mode. The procedure does not actually collect
statistics, but reports all objects that would be
affected by invoking
GATHER_DICTIONARY_STATS.
REPORT_GATHER_DATABASE_STATS Runs GATHER_DATABASE_STATS in reporting
mode. The procedure does not actually collect
statistics, but reports all objects that would be
affected by invoking GATHER_DATABASE_STATS.
REPORT_GATHER_FIXED_OBJ_STATS Runs GATHER_FIXED_OBJ_STATS in reporting
mode. The procedure does not actually collect
statistics, but reports all objects that would be
affected by invoking GATHER_FIXED_OBJ_STATS.
REPORT_GATHER_AUTO_STATS Runs the automatic statistics gather job in
reporting mode. The procedure does not actually
collect statistics, but reports all objects that would
be affected by running the job.
Assumptions
This tutorial assumes that you want to generate an HTML report of the objects that
would be affected by running GATHER_SCHEMA_STATS on the oe schema.
13-46
Chapter 13
Running Statistics Gathering Functions in Reporting Mode
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS
13-47
14
Managing Extended Statistics
DBMS_STATS enables you to collect extended statistics, which are statistics that can
improve cardinality estimates when multiple predicates exist on different columns of a
table, or when predicates use expressions.
An extension is either a column group or an expression. Column group statistics can
improve cardinality estimates when multiple columns from the same table occur
together in a SQL statement. Expression statistics improves optimizer estimates when
predicates use expressions, for example, built-in or user-defined functions.
Note:
You cannot create extended statistics on virtual columns.
See Also:
Oracle Database SQL Language Reference for a list of restrictions on virtual
columns
14-1
Chapter 14
Managing Column Group Statistics
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS package
When the WHERE clause includes multiple predicates on different columns from the
same table, individual column statistics do not show the relationship between the
columns. This is the problem solved by a column group.
The optimizer calculates the selectivity of the predicates independently, and then
combines them. However, if a correlation between the individual columns exists, then
the optimizer cannot take it into account when determining a cardinality estimate,
which it creates by multiplying the selectivity of each table predicate by the number of
rows.
The following graphic contrasts two ways of gathering statistics on the
cust_state_province and country_id columns of the sh.customers table. The
diagram shows DBMS_STATS collecting statistics on each column individually and on the
group. The column group has a system-generated name.
14-2
Chapter 14
Managing Column Group Statistics
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_
Statistics for
Column Group
DBMS_STATS
Note:
The optimizer uses column group statistics for equality predicates, inlist
predicates, and for estimating the GROUP BY cardinality.
14-3
Chapter 14
Managing Column Group Statistics
SELECT COUNT(*)
FROM sh.customers
WHERE cust_state_province = 'CA';
COUNT(*)
----------
3341
Consider an explain plan for a query of customers in the state CA and in the country
with ID 52790 (USA):
Explained.
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
Plan hash value: 1683234692
---------------------------------------------------------------------------
----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT | | 128 | 24192 | 442 (7)|
00:00:06 |
|* 1 | TABLE ACCESS FULL| CUSTOMERS | 128 | 24192 | 442 (7)|
00:00:06 |
---------------------------------------------------------------------------
----
14-4
Chapter 14
Managing Column Group Statistics
---------------------------------------------------
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
-----
13 rows selected.
See Also:
When you want to manage column group statistics manually, then use DBMS_STATS as
follows:
• Detect column groups
• Create previously detected column groups
• Create column groups manually and gather column group statistics
14-5
Chapter 14
Managing Column Group Statistics
See Also:
14-6
Chapter 14
Managing Column Group Statistics
See Also:
Assumptions
This tutorial assumes the following:
• Cardinality estimates have been incorrect for queries of the sh.customers_test
table (created from the customers table) that use predicates referencing the
columns country_id and cust_state_province.
• You want the database to monitor your workload for 5 minutes (300 seconds).
• You want the database to determine which column groups are needed
automatically.
BEGIN
DBMS_STATS.SEED_COL_USAGE(null,null,300);
END;
/
4. As user sh, run explain plans for two queries in the workload.
The following examples show the explain plans for two queries on the
customers_test table:
14-7
Chapter 14
Managing Column Group Statistics
FROM customers_test
WHERE cust_city = 'Los Angeles'
AND cust_state_province = 'CA'
AND country_id = 52790;
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY('plan_table', null,'basic rows'));
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY('plan_table', null,'basic rows'));
PLAN_TABLE_OUTPUT
------------------------------------------------------------------------
---
Plan hash value: 4115398853
----------------------------------------------------
| Id | Operation | Name | Rows |
----------------------------------------------------
| 0 | SELECT STATEMENT | | 1 |
| 1 | TABLE ACCESS FULL| CUSTOMERS_TEST | 1 |
----------------------------------------------------
8 rows selected.
PLAN_TABLE_OUTPUT
------------------------------------------------------------------------
---
Plan hash value: 3050654408
-----------------------------------------------------
| Id | Operation | Name | Rows |
-----------------------------------------------------
| 0 | SELECT STATEMENT | | 1949 |
| 1 | HASH GROUP BY | | 1949 |
| 2 | TABLE ACCESS FULL| CUSTOMERS_TEST | 55500 |
-----------------------------------------------------
9 rows selected.
The first plan shows a cardinality of 1 row for a query that returns 932 rows. The
second plan shows a cardinality of 1949 rows for a query that returns 145 rows.
5. Optionally, review the column usage information recorded for the table.
14-8
Chapter 14
Managing Column Group Statistics
LEGEND:
.......
########################################################################
###
1. COUNTRY_ID : EQ
2. CUST_CITY : EQ
3. CUST_STATE_PROVINCE : EQ
4. (CUST_CITY, CUST_STATE_PROVINCE,
COUNTRY_ID) : FILTER
5. (CUST_STATE_PROVINCE, COUNTRY_ID) : GROUP_BY
########################################################################
###
In the preceding report, the first three columns were used in equality predicates in
the first monitored query:
...
WHERE cust_city = 'Los Angeles'
AND cust_state_province = 'CA'
AND country_id = 52790;
All three columns appeared in the same WHERE clause, so the report shows them
as a group filter. In the second query, two columns appeared in the GROUP BY
clause, so the report labels them as GROUP_BY. The sets of columns in the FILTER
and GROUP_BY report are candidates for column groups.
14-9
Chapter 14
Managing Column Group Statistics
See Also:
Assumptions
This tutorial assumes that you have performed the steps in "Detecting Useful Column
Groups for a Specific Workload".
########################################################################
###
EXTENSIONS FOR SH.CUSTOMERS_TEST
................................
1. (CUST_CITY, CUST_STATE_PROVINCE,
COUNTRY_ID) :SYS_STUMZ$C3AIHLPBROI#SKA58H_N
created
2. (CUST_STATE_PROVINCE, COUNTRY_ID):SYS_STU#S#WF25Z#QAHIHE#MOFFMM_
created
########################################################################
###
The database created two column groups for customers_test: one column group
for the filter predicate and one group for the GROUP BY operation.
2. Regather table statistics.
Run GATHER_TABLE_STATS to regather the statistics for customers_test:
EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'customers_test');
3. As user sh, run explain plans for two queries in the workload.
14-10
Chapter 14
Managing Column Group Statistics
This example shows the two column group names returned from the
DBMS_STATS.CREATE_EXTENDED_STATS function. The column group created on
CUST_CITY, CUST_STATE_PROVINCE, and COUNTRY_ID has a height-balanced
histogram.
4. Explain the plans again.
The following examples show the explain plans for two queries on the
customers_test table:
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY('plan_table', null,'basic rows'));
SELECT PLAN_TABLE_OUTPUT
FROM TABLE(DBMS_XPLAN.DISPLAY('plan_table', null,'basic rows'));
----------------------------------------------------
| Id | Operation | Name | Rows |
----------------------------------------------------
| 0 | SELECT STATEMENT | | 1093 |
| 1 | TABLE ACCESS FULL| CUSTOMERS_TEST | 1093 |
----------------------------------------------------
8 rows selected.
14-11
Chapter 14
Managing Column Group Statistics
-----------------------------------------------------
| Id | Operation | Name | Rows |
-----------------------------------------------------
| 0 | SELECT STATEMENT | | 145 |
| 1 | HASH GROUP BY | | 145 |
| 2 | TABLE ACCESS FULL| CUSTOMERS_TEST | 55500 |
-----------------------------------------------------
9 rows selected.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS package
Assumptions
This tutorial assumes the following:
• You want to create a column group for the cust_state_province and country_id
columns in the customers table in sh schema.
• You want to gather statistics (including histograms) on the entire table and the new
column group.
BEGIN
DBMS_STATS.GATHER_TABLE_STATS( 'sh','customers',
METHOD_OPT => 'FOR ALL COLUMNS SIZE SKEWONLY ' ||
'FOR COLUMNS SIZE SKEWONLY (cust_state_province,country_id)' );
END;
/
14-12
Chapter 14
Managing Column Group Statistics
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GATHER_TABLE_STATS procedure
You can also use views to obtain information such as the number of distinct values,
and whether the column group has a histogram.
Assumptions
This tutorial assumes the following:
• You created a column group for the cust_state_province and country_id
columns in the customers table in sh schema.
• You want to determine the column group name, the number of distinct values, and
whether a histogram has been created for a column group.
COL_GROUP_NAME
----------------
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_
EXTENSION_NAME EXTENSION
--------------------------------------------------------------------
---
14-13
Chapter 14
Managing Column Group Statistics
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_
("CUST_STATE_PROVINCE","COUNTRY_ID")
3. Query the number of distinct values and find whether a histogram has been
created for a column group.
For example, run the following query:
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.SHOW_EXTENDED_STATS_NAME function
Assumptions
This tutorial assumes the following:
• You created a column group for the cust_state_province and country_id
columns in the customers table in sh schema.
• You want to drop the column group.
BEGIN
DBMS_STATS.DROP_EXTENDED_STATS( 'sh', 'customers',
'(cust_state_province, country_id)' );
END;
/
14-14
Chapter 14
Managing Expression Statistics
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.DROP_EXTENDED_STATS function
The following graphic shows the optimizer using statistics to generate a plan for a
query that uses a function. The top shows the optimizer checking statistics for the
column. The bottom shows the optimizer checking statistics corresponding to the
expression used in the query. The expression statistics yield more accurate estimates.
14-15
Chapter 14
Managing Expression Statistics
Optimizer
Do
Use Yes expression No Use Default
Expression statistics Column
Statistics exist? Statistics
LOWER(cust_state_province) cust_state_province
Expression Statistics Column Statistics
Optimal Suboptimal
Estimate Estimate
As shown in Figure 14-2, when expression statistics are not available, the optimizer
can produce suboptimal plans.
This section contains the following topics:
See Also:
Oracle Database SQL Language Reference to learn about SQL functions
COUNT(*)
----------
3341
14-16
Chapter 14
Managing Expression Statistics
Consider the plan for the same query with the LOWER() function applied:
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------
----
Plan hash value: 2008213504
---------------------------------------------------------------------------
----
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------
----
| 0 | SELECT STATEMENT | | 555 | 108K| 406 (1)|
00:00:05 |
|* 1 | TABLE ACCESS FULL| CUSTOMERS | 555 | 108K| 406 (1)|
00:00:05 |
---------------------------------------------------------------------------
----
1 - filter(LOWER("CUST_STATE_PROVINCE")='ca')
Assumptions
This tutorial assumes the following:
• Selectivity estimates are inaccurate for queries of sh.customers that use the
UPPER(cust_state_province) function.
• You want to gather statistics on the UPPER(cust_state_province) expression.
14-17
Chapter 14
Managing Expression Statistics
BEGIN
DBMS_STATS.GATHER_TABLE_STATS(
'sh'
, 'customers'
, method_opt => 'FOR ALL COLUMNS SIZE SKEWONLY ' ||
'FOR COLUMNS (LOWER(cust_state_province)) SIZE
SKEWONLY'
);
END;
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GATHER_TABLE_STATS procedure
You can also use views to obtain information such as the number of distinct values,
and whether the column group has a histogram.
Assumptions
This tutorial assumes the following:
• You created extended statistics for the LOWER(cust_state_province) expression.
• You want to determine the column group name, the number of distinct values, and
whether a histogram has been created for a column group.
14-18
Chapter 14
Managing Expression Statistics
FROM USER_STAT_EXTENSIONS
WHERE TABLE_NAME='CUSTOMERS';
EXTENSION_NAME EXTENSION
------------------------------ ------------------------------
SYS_STUBPHJSBRKOIK9O2YV3W8HOUE (LOWER("CUST_STATE_PROVINCE"))
3. Query the number of distinct values and find whether a histogram has been
created for the expression.
For example, run the following query:
See Also:
Assumptions
This tutorial assumes the following:
• You created extended statistics for the LOWER(cust_state_province) expression.
• You want to drop the expression statistics.
14-19
Chapter 14
Managing Expression Statistics
For example, the following PL/SQL program deletes a column group from the
customers table:
BEGIN
DBMS_STATS.DROP_EXTENDED_STATS(
'sh'
, 'customers'
, '(LOWER(cust_state_province))'
);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.DROP_EXTENDED_STATS procedure
14-20
15
Controlling the Use of Optimizer Statistics
Using DBMS_STATS, you can specify when and how the optimizer uses statistics.
When you lock statistics on a table, all dependent statistics are locked. The locked
statistics include table statistics, column statistics, histograms, and dependent index
statistics. To overwrite statistics even when they are locked, you can set the value of
the FORCE argument in various DBMS_STATS procedures, for example, DELETE_*_STATS
and RESTORE_*_STATS, to true.
Assumptions
This tutorial assumes the following:
15-1
Chapter 15
Locking and Unlocking Optimizer Statistics
To lock statistics:
1. Start SQL*Plus and connect to the database as the oe user.
2. Lock the statistics on oe.orders.
For example, execute the following PL/SQL program:
BEGIN
DBMS_STATS.LOCK_TABLE_STATS('OE','ORDERS');
END;
/
BEGIN
DBMS_STATS.LOCK_SCHEMA_STATS('HR');
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.LOCK_TABLE_STATS procedure
Assumptions
This tutorial assumes the following:
• You locked statistics on the oe.orders table and on the hr schema.
• You want to unlock these statistics.
To unlock statistics:
1. Start SQL*Plus and connect to the database as the oe user.
2. Unlock the statistics on oe.orders.
15-2
Chapter 15
Publishing Pending Optimizer Statistics
BEGIN
DBMS_STATS.UNLOCK_TABLE_STATS('OE','ORDERS');
END;
/
BEGIN
DBMS_STATS.UNLOCK_SCHEMA_STATS('HR');
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.UNLOCK_TABLE_STATS procedure
15-3
Chapter 15
Publishing Pending Optimizer Statistics
By default, the optimizer uses published statistics. You can change the default
behavior by setting the OPTIMIZER_USE_PENDING_STATISTICS initialization parameter to
true (the default is false).
The top part of the following graphic shows the optimizer gathering statistics for the
sh.customers table and storing them in the data dictionary with pending status. The
bottom part of the diagram shows the optimizer using only published statistics to
process a query of sh.customers.
Data Dictionary
Published
1 0 0 1 1 1 Statistics
0 1 0 0 0 1
Optimizer
0 0 1 0 0 0 Pending
Statistics
1 1 0 0 1 0
Publishing
preferences
set to false
Customers
Table
Data Dictionary
Optimizer Statistics
Optimizer Published
SELECT ... 1 0 0 1 1 1 Statistics
FROM 0 1 0 0 0 1
customers
0 0 1 0 0 0 Pending
Statistics
1 1 0 0 1 0
OPTIMIZER_USE_PENDING_STATISTICS=false
In some cases, the optimizer can use a combination of published and pending
statistics. For example, the database stores both published and pending statistics for
the customers table. For the orders table, the database stores only published
statistics. If OPTIMIZER_USE_PENDING_STATS = true, then the optimizer uses pending
statistics for customers and published statistics for orders. If
OPTIMIZER_USE_PENDING_STATS = false, then the optimizer uses published statistics
for customers and orders.
15-4
Chapter 15
Publishing Pending Optimizer Statistics
See Also:
Oracle Database Reference to learn about the
OPTIMIZER_USE_PENDING_STATISTICS initialization parameter
View Description
USER_TAB_STATISTICS Displays optimizer statistics for the tables accessible
to the current user.
USER_TAB_COL_STATISTICS Displays column statistics and histogram information
extracted from ALL_TAB_COLUMNS.
15-5
Chapter 15
Publishing Pending Optimizer Statistics
View Description
USER_PART_COL_STATISTICS Displays column statistics and histogram information
for the table partitions owned by the current user.
USER_SUBPART_COL_STATISTICS Describes column statistics and histogram
information for subpartitions of partitioned objects
owned by the current user.
USER_IND_STATISTICS Displays optimizer statistics for the indexes
accessible to the current user.
USER_TAB_PENDING_STATS Describes pending statistics for tables, partitions, and
subpartitions accessible to the current user.
USER_COL_PENDING_STATS Describes the pending statistics of the columns
accessible to the current user.
USER_IND_PENDING_STATS Describes the pending statistics for tables, partitions,
and subpartitions accessible to the current user
collected using the DBMS_STATS package.
See Also:
Assumptions
This tutorial assumes the following:
• You want to change the preferences for the sh.customers and sh.sales tables so
that newly collected statistics have pending status.
• You want the current session to use pending statistics.
• You want to gather and publish pending statistics on the sh.customers table.
• You gather the pending statistics on the sh.sales table, but decide to delete them
without publishing them.
• You want to change the preferences for the sh.customers and sh.sales tables so
that newly collected statistics are published.
15-6
Chapter 15
Publishing Pending Optimizer Statistics
PUBLISH
-------
TRUE
The value true indicates that the database publishes statistics as it gathers them.
Every table uses this value unless a specific table preference has been set.
When using GET_PREFS, you can also specify a schema and table name. The
function returns a table preference if it is set. Otherwise, the function returns the
global preference.
3. Query the pending statistics.
For example, run the following query (sample output included):
no rows selected
This example shows that the database currently stores no pending statistics for
the sh schema.
4. Change the publishing preferences for the sh.customers table.
For example, execute the following procedure so that statistics are marked as
pending:
BEGIN
DBMS_STATS.SET_TABLE_PREFS('sh', 'customers', 'publish', 'false');
END;
/
Subsequently, when you gather statistics on the customers table, the database
does not automatically publish statistics when the gather job completes. Instead,
the database stores the newly gathered statistics in the USER_TAB_PENDING_STATS
table.
5. Gather statistics for sh.customers.
For example, run the following program:
BEGIN
DBMS_STATS.GATHER_TABLE_STATS('sh','customers');
END;
/
15-7
Chapter 15
Publishing Pending Optimizer Statistics
TABLE_NAME NUM_ROWS
------------------------------ ----------
CUSTOMERS 55500
This example shows that the database now stores pending statistics for the
sh.customers table.
7. Instruct the optimizer to use the pending statistics in this session.
Set the initialization parameter OPTIMIZER_USE_PENDING_STATISTICS to true as
shown:
8. Run a workload.
The following example changes the email addresses of all customers named
Bruce Chalmers:
UPDATE sh.customers
SET cust_email='ChalmersB@company.example.icom'
WHERE cust_first_name = 'Bruce'
AND cust_last_name = 'Chalmers';
COMMIT;
The optimizer uses the pending statistics instead of the published statistics when
compiling all SQL statements in this session.
9. Publish the pending statistics for sh.customers.
For example, execute the following program:
BEGIN
DBMS_STATS.PUBLISH_PENDING_STATS('SH','CUSTOMERS');
END;
/
BEGIN
DBMS_STATS.SET_TABLE_PREFS('sh', 'sales', 'publish', 'false');
END;
/
Subsequently, when you gather statistics on the sh.sales table, the database
does not automatically publish statistics when the gather job completes. Instead,
the database stores the statistics in the USER_TAB_PENDING_STATS table.
11. Gather statistics for sh.sales.
15-8
Chapter 15
Creating Artificial Optimizer Statistics for Testing
BEGIN
DBMS_STATS.GATHER_TABLE_STATS('sh','sales');
END;
/
Assume you change your mind and now want to delete pending statistics for
sh.sales. Run the following program:
BEGIN
DBMS_STATS.DELETE_PENDING_STATS('sh','sales');
END;
/
13. Change the publishing preferences for the sh.customers and sh.sales tables
back to their default setting.
For example, execute the following program:
BEGIN
DBMS_STATS.SET_TABLE_PREFS('sh', 'customers', 'publish', null);
DBMS_STATS.SET_TABLE_PREFS('sh', 'sales', 'publish', null);
END;
/
15-9
Chapter 15
Creating Artificial Optimizer Statistics for Testing
Caution:
The DBMS_STATS.SET_*_STATS procedures are intended for development
testing only. Do not use them in a production database. If you set statistics in
the data dictionary, then Oracle Database considers the set statistics as the
“real” statistics, which means that statistics gathering jobs may not re-gather
artificial statistics when they do not meet the criteria for staleness.
• Showing how execution plans change as the numbers of rows or blocks in a table
change
For example, SET_TABLE_STATS can set number of rows and blocks in a small or
empty table to a large number. When you execute a query using the altered
statistics, the optimizer may change the execution plan. For example, the
increased row count may lead the optimizer to choose an index scan rather than a
full table scan. By experimenting with different values, you can see how the
optimizer will change its execution plan over time.
• Creating realistic statistics for temporary tables
You may want to see what the optimizer does when a large temporary table is
referenced in multiple SQL statements. You can create a regular table, load
representative data, and then use GET_TABLE_STATS to retrieve the statistics. After
you create the temporary table, you can “deceive” the optimizer into using these
statistics by invoking SET_TABLE_STATS.
Optionally, you can specify a unique ID for statistics in a user-created table. The
SET_*_STATS procedures have corresponding GET_*_STATS procedures.
15-10
Chapter 15
Creating Artificial Optimizer Statistics for Testing
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.SET_TABLE_STATS and the other procedures for setting
optimizer statistics
15-11
Chapter 15
Creating Artificial Optimizer Statistics for Testing
15-12
Chapter 15
Creating Artificial Optimizer Statistics for Testing
4. Query the number of rows for the table and index (sample output included):
NUM_ROWS
----------
1
NUM_ROWS
----------
1
6. Query the execution plan chosen by the optimizer (sample output included):
----------------------------------------------------------------------
| Id | Operation | Name |Rows|Bytes|Cost (%CPU)| Time|
----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 2 (100)| |
|* 1 | TABLE ACCESS FULL| CONTRACTORS | 1 | 12 | 2 (0)| 00:00:01 |
----------------------------------------------------------------------
1 - filter("SALARY"=1000)
19 rows selected.
Because only 1 row exists in the table, the optimizer chooses a full table scan over
an index range scan.
7. Use SET_TABLE_STATS and SET_INDEX_STATS to simulate statistics for a table with
2000 rows stored in 10 data blocks:
BEGIN
DBMS_STATS.SET_TABLE_STATS(
15-13
Chapter 15
Creating Artificial Optimizer Statistics for Testing
BEGIN
DBMS_STATS.SET_INDEX_STATS(
ownname => user
, indname => 'SALARY_IX'
, numrows => 2000 );
END;
/
8. Query the number of rows for the table and index (sample output included):
NUM_ROWS
----------
2000
NUM_ROWS
----------
2000
Now the optimizer believes that the table contains 2000 rows in 10 blocks, even
though only 1 row actually exists in one block.
9. Flush the shared pool to eliminate possibility of plan reuse, and then execute the
same query of contractors:
10. Query the execution plan chosen by the optimizer based on the artificial statistics
(sample output included):
------------------------------------------------------------------------
15-14
Chapter 15
Creating Artificial Optimizer Statistics for Testing
---------
|Id| Operation | Name |Rows|Bytes|
Cost(%CPU)|Time|
------------------------------------------------------------------------
---------
| 0| SELECT STATEMENT | | | |
3(100)| |
| 1| TABLE ACCESS BY INDEX ROWID BATCHED|CONTRACTORS|2000|24000|3 (34)|
00:00:01|
|*2| INDEX RANGE SCAN |SALARY_IX |2000| |1 (0)|
00:00:01|
------------------------------------------------------------------------
---------
2 - access("SALARY"=1000)
20 rows selected.
Based on the artificially generated statistics for the number of rows and block
distribution, the optimizer considers an index range scan more cost-effective.
15-15
16
Managing Historical Optimizer Statistics
This chapter how to retain, report on, and restore non-current statistics.
This chapter contains the following topics:
• Restoring Optimizer Statistics
You can use DBMS_STATS to restore old versions of statistics that are stored in the
data dictionary.
• Managing Optimizer Statistics Retention
By default, the database retains optimizer statistics for 31 days, after which time
the statistics are scheduled for purging.
• Reporting on Past Statistics Gathering Operations
You can use DBMS_STATS functions to report on a specific statistics gathering
operation or on operations that occurred during a specified time.
16-1
Chapter 16
Restoring Optimizer Statistics
See Also:
Oracle Database PL/SQL Packages and Types Reference for an overview of
the procedures for restoring and importing statistics
16-2
Chapter 16
Restoring Optimizer Statistics
The procedures listed in the following table accept a timestamp as an argument and
restore statistics as of the specified time (as_of_timestamp).
Procedure Description
RESTORE_DICTIONARY_STATS Restores statistics of all dictionary tables (tables of
SYS, SYSTEM, and RDBMS component schemas) as of
a specified timestamp.
RESTORE_FIXED_OBJECTS_STATS Restores statistics of all fixed tables as of a specified
timestamp.
RESTORE_SCHEMA_STATS Restores statistics of all tables of a schema as of a
specified timestamp.
RESTORE_SYSTEM_STATS Restores system statistics as of a specified
timestamp.
RESTORE_TABLE_STATS Restores statistics of a table as of a specified
timestamp. The procedure also restores statistics of
associated indexes and columns. If the table statistics
were locked at the specified timestamp, then the
procedure locks the statistics.
Dictionary views display the time of statistics modifications. You can use the following
views to determine the time stamp to be use for the restore operation:
• The DBA_OPTSTAT_OPERATIONS view contain history of statistics operations
performed at schema and database level using DBMS_STATS.
• The DBA_TAB_STATS_HISTORY views contains a history of table statistics
modifications.
Assumptions
This tutorial assumes the following:
• After the most recent statistics collection for the oe.orders table, the optimizer
began choosing suboptimal plans for queries of this table.
• You want to restore the statistics from before the most recent statistics collection
to see if the plans improve.
16-3
Chapter 16
Managing Optimizer Statistics Retention
STATS_MOD_TIME
FROM DBA_TAB_STATS_HISTORY
WHERE TABLE_NAME='ORDERS'
AND OWNER='OE'
ORDER BY STATS_UPDATE_TIME DESC;
TABLE_NAME STATS_MOD_TIME
---------- -------------------
ORDERS 2012-08-20:11:36:38
ORDERS 2012-08-10:11:06:20
BEGIN
DBMS_STATS.RESTORE_TABLE_STATS( 'OE','ORDERS',
TO_TIMESTAMP('2012-08-10:11:06:20','YYYY-MM-
DD:HH24:MI:SS') );
END;
/
You can specify any date between 8/10 and 8/20 because DBMS_STATS restores
statistics as of the specified time.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DBMS_STATS.RESTORE_TABLE_STATS procedure
16-4
Chapter 16
Managing Optimizer Statistics Retention
DECLARE
v_stats_retn NUMBER;
v_stats_date DATE;
BEGIN
v_stats_retn := DBMS_STATS.GET_STATS_HISTORY_RETENTION;
DBMS_OUTPUT.PUT_LINE('The retention setting is ' || v_stats_retn ||
'.');
v_stats_date := DBMS_STATS.GET_STATS_HISTORY_AVAILABILITY;
DBMS_OUTPUT.PUT_LINE('Earliest restore date is ' || v_stats_date ||
'.');
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.GET_STATS_HISTORY_RETENTION procedure
16-5
Chapter 16
Managing Optimizer Statistics Retention
Prerequisites
To run this procedure, you must have either the SYSDBA privilege, or both the ANALYZE
ANY DICTIONARY and ANALYZE ANY system privileges.
Assumptions
This tutorial assumes the following:
• The current retention period for optimizer statistics is 31 days.
• You run queries annually as part of an annual report. To keep the statistics history
for more than 365 days so that you have access to last year's plan (in case a
suboptimal plan occurs now), you set the retention period to 366 days.
• You want to create a PL/SQL procedure set_opt_stats_retention that you can
use to change the optimizer statistics retention period.
16-6
Chapter 16
Reporting on Past Statistics Gathering Operations
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.ALTER_STATS_HISTORY_RETENTION procedure
The database purges all history older than the older of (current time - the
ALTER_STATS_HISTORY_RETENTION setting) and (time of the most recent statistics
gathering - 1).
You can purge old statistics manually using the PURGE_STATS procedure. If you do not
specify an argument, then this procedure uses the automatic purging policy. If you
specify the before_timestamp parameter, then the database purges statistics saved
before the specified timestamp.
Prerequisites
To run this procedure, you must have either the SYSDBA privilege, or both the ANALYZE
ANY DICTIONARY and ANALYZE ANY system privileges.
Assumptions
This tutorial assumes that you want to purge statistics more than one week old.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS.PURGE_STATS procedure
16-7
Chapter 16
Reporting on Past Statistics Gathering Operations
Function Description
REPORT_STATS_OPERATIONS Generates a report of all statistics operations that
occurred between two points in time. You can
narrow the scope of the report to include only
automatic statistics gathering runs. You can also
provide a set of pluggable database (PDB) IDs so
that the database reports only statistics operations
from the specified PDBs.
REPORT_SINGLE_STATS_OPERATION Generates a report of the specified operation.
Optionally, you can specify a particular PDB ID in a
container database (CDB).
Assumptions
This tutorial assumes that you want to generate HTML reports of the following:
• All statistics gathering operations within the last day
• The most recent statistics gathering operation
16-8
Chapter 16
Reporting on Past Statistics Gathering Operations
BEGIN
:my_report :=DBMS_STATS.REPORT_SINGLE_STATS_OPERATION (
OPID => 848
, FORMAT => 'HTML'
);
END;
See Also:
16-9
17
Importing and Exporting Optimizer
Statistics
You can export and import optimizer statistics from the data dictionary to user-defined
statistics tables. You can also copy statistics from one database to another database.
This chapter contains the following topics:
• About Transporting Optimizer Statistics
When you transport optimizer statistics between databases, you must use
DBMS_STATS to copy the statistics to and from a staging table, and tools to make
the table contents accessible to the destination database.
• Transporting Optimizer Statistics to a Test Database: Tutorial
You can transport schema statistics from a production database to a test database
using Oracle Data Pump.
Production Test
Database Database
17-1
Chapter 17
Transporting Optimizer Statistics to a Test Database: Tutorial
Note:
Exporting and importing statistics using DBMS_STATS is a distinct
operation from using Data Pump Export and Import.
Assumptions
This tutorial assumes the following:
• You want to generate representative sh schema statistics on a production
database and use DBMS_STATS to import them into a test database.
• Administrative user dba1 exists on both production and test databases.
• You intend to create table opt_stats to store the schema statistics.
• You intend to use Oracle Data Pump to export and import table opt_stats.
17-2
Chapter 17
Transporting Optimizer Statistics to a Test Database: Tutorial
BEGIN
DBMS_STATS.CREATE_STAT_TABLE (
ownname => 'dba1'
, stattab => 'opt_stats'
);
END;
/
BEGIN
DBMS_STATS.EXPORT_SCHEMA_STATS (
ownname => 'dba1'
, stattab => 'opt_stats'
);
END;
/
5. Use Oracle Data Pump to export the contents of the statistics table.
For example, run the expdp command at the operating schema prompt:
8. On the test host, start SQL*Plus and connect to the test database as administrator
dba1.
17-3
Chapter 17
Transporting Optimizer Statistics to a Test Database: Tutorial
9. Use DBMS_STATS to import statistics from the user statistics table and store them in
the data dictionary.
The following PL/SQL program imports schema statistics from table opt_stats
into the data dictionary:
BEGIN
DBMS_STATS.IMPORT_SCHEMA_STATS(
ownname => 'dba1'
, stattab => 'opt_stats'
);
END;
/
See Also:
17-4
18
Analyzing Statistics Using Optimizer
Statistics Advisor
Optimizer Statistics Advisor analyzes how optimizer statistics are gathered, and then
makes recommendations.
This chapter contains the following topics:
• About Optimizer Statistics Advisor
Optimizer Statistics Advisor is built-in diagnostic software that analyzes the quality
of statistics and statistics-related tasks.
• Basic Tasks for Optimizer Statistics Advisor
This section explains the basic workflow for using Optimizer Statistics Advisor. All
procedures and functions are in the DBMS_STATS package.
18-1
Chapter 18
About Optimizer Statistics Advisor
Optimizer
Automatic
Tuning
Optimizer
Filter Options
DBA Optimizer Findings
Statistics
Advisor Recommendations
Task
Actions
Rules
Data Dictionary
and V$ Views
DBA_OPSTAT_OPERATIONS
18-2
Chapter 18
About Optimizer Statistics Advisor
Optimizer statistics play a significant part in determining the execution plan for queries.
Therefore, it is critical for the optimizer to gather and maintain accurate and up-to-date
statistics. The optimizer provides the DBMS_STATS package, which evolves from release
to release, for this purpose. Typically, users develop their own strategies for gathering
statistics based on specific workloads, and then use homegrown scripts to implement
these strategies.
This section contains the following topics:
• Problems with a Traditional Script-Based Approach
The advantage of the scripted approach is that the scripts are typically tested and
reviewed. However, the owner of suboptimal legacy scripts may not change them
for fear of causing plan changes.
• Advantages of Optimizer Statistics Advisor
An advisor-based approach offers better scalability and maintainability than the
traditional approach.
18-3
Chapter 18
About Optimizer Statistics Advisor
great deal of time emailing scripts of the problematic queries, enabling traces, and
investigating traces.
• Recommended fixes may not be feasible.
Performance engineers may recommend changing the application code that
maintains statistics. In some organizations, this requirement may be difficult or
impossible to satisfy.
18-4
Chapter 18
About Optimizer Statistics Advisor
Rules
Findings
Recommendations
Actions
18-5
Chapter 18
About Optimizer Statistics Advisor
The rules embody Oracle best practices based on the current feature set. If the best
practices change from release to release, then the Optimizer Statistics Advisor rules
also change.
The advisor organizes rules into the following classes:
• System
This class checks the preferences for statistics collection, status of the automated
statistics gathering job, use of SQL plan directives, and so on. Rules in this class
have the value SYSTEM in V$STATS_ADVISOR_RULES.RULE_TYPE.
• Operation
This class checks whether statistics collection uses the defaults, test statistics are
created using the SET_*_STATS procedures, and so on. Rules in this class have the
value OPERATION in V$STATS_ADVISOR_RULES.RULE_TYPE.
• Object
This class checks for the quality of the statistics, staleness of statistics,
unnecessary collection of statistics, and so on. Rules in this class have the value
OBJECT in V$STATS_ADVISOR_RULES.RULE_TYPE.
The rules check for the following problems:
• How to gather statistics
For example, one rule might specify the recommended setting for an initialization
parameter. Another rule might specify that statistics should be gathered at the
schema level.
• When to gather statistics
For example, the advisor may recommend that the maintenance window for the
automatic statistics gathering job should be enabled, or that the window should be
extended.
• How to improve the efficiency of statistics gathering
For example, a rule might specify that default parameters should be used in
DBMS_STATS, or that statistics should not be set manually.
In V$STATS_ADVISOR_RULES, each rule has a unique string ID that is usable in the
DBMS_STATS procedures and reports. You can use a rule filter to specify rules that
Optimizer Statistics Advisor should check. However, you cannot write new rules.
Example 18-1 Listing Rules in V$STATS_ADVISOR_RULES
The following query, with sample output, lists a subset of the rules in
V$STATS_ADVISOR_RULES. The rules may change from release to release.
18-6
Chapter 18
About Optimizer Statistics Advisor
12 rows selected.
See Also:
Oracle Database Reference to learn more about V$STATS_ADVISOR_RULES
A finding corresponds to exactly one rule. However, a rule can generate many
findings.
18-7
Chapter 18
About Optimizer Statistics Advisor
See Also:
Multiple recommendations may exist for a single finding. In this case, you must
investigate to determine which recommendation to follow. Each recommendation
includes one or more rationales that explain why Optimizer Statistics Advisor makes its
recommendation. In some cases, findings may not generate recommendations.
See Also:
• "Guideline for Setting the Sample Size" to learn the guideline for the
sample size
• Oracle Database Reference to learn about
DBA_ADVISOR_RECOMMENDATIONS
For example, Optimizer Statistics Advisor executes a task that performs the following
steps:
1. Checks rules
The advisor checks conformity to the rule that stale statistics should be avoided.
2. Generates finding
The advisor discovers that a number of objects have no statistics.
3. Generates recommendation
The advisor recommends gathering statistics on the objects with no statistics.
4. Generates action
18-8
Chapter 18
About Optimizer Statistics Advisor
See Also:
See Also:
18-9
Chapter 18
About Optimizer Statistics Advisor
18-10
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Table 18-4 (Cont.) DBMS_STATS APIs for Task and Filter Configuration
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS package
The following figure shows the automatic and manual paths in the workflow. If
AUTO_STATS_ADVISOR_TASK runs automatically in the maintenance window, then your
workflow begins by querying the report. In the manual workflow, you must use PL/SQL
to create and execute the tasks.
18-11
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
CONFIGURE_ADVISOR_*_FILTER Optionally,
alter the
scope of
the advisor
Execute the advisor checks
task EXECUTE_ADVISOR_TASK
Generate a report of
findings and REPORT_ADVISOR_TASK
recommendations
Generate
modifiable
PL/SQL
IMPLEMENT_ADVISOR_TASK SCRIPT_ADVISOR_TASK script
Run PL/SQL
script
Typically, you perform Optimizer Statistics Advisor steps in the sequence shown in the
following table.
18-12
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Example 18-2 Basic Script for Optimizer Statistics Advisor in Manual Workflow
This script illustrates a basic Optimizer Statistics Advisor session. It creates a task,
executes it, generates a report, and then implements the recommendations.
DECLARE
v_tname VARCHAR2(128) := 'my_task';
v_ename VARCHAR2(128) := NULL;
v_report CLOB := null;
v_script CLOB := null;
v_implementation_result CLOB;
BEGIN
-- create a task
v_tname := DBMS_STATS.CREATE_ADVISOR_TASK(v_tname);
18-13
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_STATS package
Prerequisites
To execute this subprogram, you must have the ADVISOR privilege.
Note:
This subprogram executes using invoker's rights.
18-14
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
For example, to create the task opt_adv_task1, use the following code:
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.CREATE_ADVISOR_TASK(v_tname);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about CREATE_ADVISOR_TASK
18-15
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
8 rows selected.
See Also:
Oracle Database Reference to learn more about DBA_ADVISOR_EXECUTIONS
18-16
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
advantage of filters is the ability to ignore recommendations that you are not interested
in, and reduce the overhead of the advisor task.
The simplest way to create filters is to use the following DBMS_STATS procedures either
individually or in combination:
• CONFIGURE_ADVISOR_OBJ_FILTER
Use this procedure to include or exclude the specified database schemas or
objects. The object filter takes in an owner name and an object name, with
wildcards (%) supported.
• CONFIGURE_ADVISOR_RULE_FILTER
Use this procedure to include or exclude the specified rules. Obtain the names of
rules by querying V$STATS_ADVISOR_RULES.
• CONFIGURE_ADVISOR_OPR_FILTER
Use this procedure to include or exclude the specified DBMS_STATS operations.
Obtain the IDs and names for operations by querying DBA_OPTSTAT_OPERATIONS.
For the preceding functions, you can specify the type of operation to which the filter
applies: EXECUTE, REPORT, SCRIPT, and IMPLEMENT. You can also combine types, as in
EXECUTE + REPORT. Null indicates that the filter applies to all types of advisor
operations.
See Also:
Prerequisites
To use the DBMS_STATS.CONFIGURE_ADVISOR_OBJ_FILTER function, you must meet the
following prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
18-17
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Note:
This subprogram executes using invoker's rights.
BEGIN
report := DBMS_STATS.CONFIGURE_ADVISOR_OBJ_FILTER(
task_name => 'tname'
, stats_adv_opr_type => 'opr_type'
, rule_name => 'rule'
, ownname => 'owner'
, tabname => 'table'
, action => 'action' );
END;
DECLARE
v_tname VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
DBMS_STATS.DROP_ADVISOR_TASK(v_tname);
END;
/
18-18
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
v_retc := DBMS_STATS.CONFIGURE_ADVISOR_OBJ_FILTER(
task_name => p_tname
, stats_adv_opr_type => 'EXECUTE'
, rule_name => NULL
, ownname => 'SH'
, tabname => NULL
, action => 'ENABLE' );
END;
/
SHOW ERRORS
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.CREATE_ADVISOR_TASK(v_tname);
sh_obj_filter(v_tname);
END;
/
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
begin
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.EXECUTE_ADVISOR_TASK(v_tname);
END;
/
18-19
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.CONFIGURE_ADVISOR_OBJ_FILTER
Prerequisites
To use the DBMS_STATS.CONFIGURE_ADVISOR_RULE_FILTER function, you must meet the
following prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
Note:
This subprogram executes using invoker's rights.
18-20
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
BEGIN
report := DBMS_STATS.DBMS_STATS.CONFIGURE_ADVISOR_RULE_FILTER(
task_name => 'tname'
, stats_adv_opr_type => 'opr_type'
, rule_name => 'rule'
, action => 'action' );
END;
18-21
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Example 18-5 Including Only the Rule for Avoiding Stale Statistics
This example is the inverse of the preceding example. You want to generate a report
for the task named opt_adv_task1, but want to see only recommendations about stale
statistics.
1. Query V$STATS_ADVISOR_RULES for rules that deal with stale statistics (sample
output included):
BEGIN
:b_ret := DBMS_STATS.CONFIGURE_ADVISOR_RULE_FILTER(
task_name => 'opt_adv_task1'
, stats_adv_opr_type => 'EXECUTE'
, rule_name => 'AvoidStaleStats'
, action => 'ENABLE' );
END;
/
18-22
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
See Also:
Prerequisites
To use DBMS_STATS.CONFIGURE_ADVISOR_OPR_FILTER function, you must meet the
following prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
Note:
This subprogram executes using invoker's rights.
OPERATION
-----------------------
18-23
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
gather_dictionary_stats
gather_index_stats
gather_schema_stats
gather_table_stats
purge_stats
set_system_stats
SELECT ID
FROM DBA_OPTSTAT_OPERATIONS
WHERE ( OPERATION = 'gather_table_stats'
OR OPERATION = 'gather_index_stats')
AND ( TARGET LIKE 'SH.%'
OR TARGET LIKE 'SYS.%');
BEGIN
report := DBMS_STATS.CONFIGURE_ADVISOR_OPR_FILTER(
task_name => 'tname'
, stats_adv_opr_type => 'opr_type'
, rule_name => 'rule'
, operation_id => 'op_id'
, action => 'action' );
END;
18-24
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
DECLARE
v_tname VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
DBMS_STATS.DROP_ADVISOR_TASK(v_tname);
END;
/
4. Create a task named opt_adv_task1, and then execute the opr_filter procedure
for this task.
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.CREATE_ADVISOR_TASK(v_tname);
opr_filter(v_tname);
END;
/
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
begin
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.EXECUTE_ADVISOR_TASK(v_tname);
18-25
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
END;
/
SPOOL /tmp/rep.txt
SET LONG 1000000
COLUMN report FORMAT A200
SET LINESIZE 250
SET PAGESIZE 1000
SELECT DBMS_STATS.REPORT_ADVISOR_TASK(
task_name => 'opt_adv_task1'
, execution_name => NULL
, type => 'TEXT'
, section => 'ALL'
) AS report
FROM DUAL;
SPOOL OFF
See Also:
18-26
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
– Users with the ANALYZE ANY DICTIONARY privilege but not the ANALYZE ANY
privilege can perform this task for statistics operations related to their own
schema and the SYS schema.
– Users with neither the ANALYZE ANY nor the ANALYZE ANY DICTIONARY privilege
can only perform this operation for statistics operations relating to their own
schema.
• Object level
Users can perform this task for any object for which they have statistics collection
privileges.
Prerequisites
This task has the following prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
• If you specify an execution name, then this name must not conflict with an existing
execution.
Note:
This subprogram executes using invoker's rights.
For example, to execute the task opt_adv_task1, use the following code:
DECLARE
v_tname VARCHAR2(32767);
v_ret VARCHAR2(32767);
BEGIN
v_tname := 'opt_adv_task1';
v_ret := DBMS_STATS.EXECUTE_ADVISOR_TASK(v_tname);
END;
/
18-27
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about EXECUTE_ADVISOR_TASK
Prerequisites
To generate a report with the DBMS_STATS.REPORT_ADVISOR_TASK function, you must
meet the following prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
18-28
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Note:
This subprogram executes using invoker's rights.
The results of performing this task depend on the privileges of the executing user:
• SYSTEM level
Only users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges
can perform this task on system-level rules.
• Operation level
The results depend on the following privileges:
– Users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges can
perform this task for all statistics operations.
– Users with the ANALYZE ANY privilege but not the ANALYZE ANY DICTIONARY
privilege can perform this task for statistics operations related to any schema
except SYS.
– Users with the ANALYZE ANY DICTIONARY privilege but not the ANALYZE ANY
privilege can perform this task for statistics operations related to their own
schema and the SYS schema.
– Users with neither the ANALYZE ANY nor the ANALYZE ANY DICTIONARY privilege
can only perform this operation for statistics operations relating to their own
schema.
• Object level
Users can perform this task for any object for which they have statistics collection
privileges.
18-29
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
'lvl') AS REPORT
FROM DUAL;
SELECT
DBMS_STATS.REPORT_ADVISOR_TASK('AUTO_STATS_ADVISOR_TASK',NULL,'TEXT','AL
L','ALL') AS REPORT
FROM DUAL;
GENERAL INFORMATION
-------------------------------------------------------------------------------
Task Name : AUTO_STATS_ADVISOR_TASK
Execution Name : EXEC_136
Created : 09-05-16 02:52:34
Last Modified : 09-05-16 12:31:24
-------------------------------------------------------------------------------
SUMMARY
-------------------------------------------------------------------------------
For execution EXEC_136 of task AUTO_STATS_ADVISOR_TASK, the Statistics Advisor
has 4 findings. The findings are related to the following rules:
AVOIDSETPROCEDURES, USEDEFAULTPARAMS, USEGATHERSCHEMASTATS, NOTUSEINCREMENTAL.
Please refer to the finding section for detailed information.
-------------------------------------------------------------------------------
FINDINGS
-------------------------------------------------------------------------------
Rule Name: AvoidSetProcedures
Rule Description: Avoid Set Statistics Procedures
Finding: There are 5 SET_[COLUMN|INDEX|TABLE|SYSTEM]_STATS procedures being
used for statistics gathering.
Recommendation: Do not use SET_[COLUMN|INDEX|TABLE|SYSTEM]_STATS procedures.
Gather statistics instead of setting them.
Rationale: SET_[COLUMN|INDEX|TABLE|SYSTEM]_STATS will cause bad plans due to
wrong or inconsistent statistics.
----------------------------------------------------
Rule Name: UseDefaultParams
Rule Description: Use Default Parameters in Statistics Collection Procedures
Finding: There are 367 statistics operations using nondefault parameters.
Recommendation: Use default parameters for statistics operations.
Example:
-- Gathering statistics for 'SH' schema using all default parameter values:
BEGIN dbms_stats.gather_schema_stats('SH'); END;
Rationale: Using default parameter values for statistics gathering operations
is more efficient.
----------------------------------------------------
Rule Name: UseGatherSchemaStats
Rule Description: Use gather_schema_stats procedure
Finding: There are 318 uses of GATHER_TABLE_STATS.
Recommendation: Use GATHER_SCHEMA_STATS instead of GATHER_TABLE_STATS.
18-30
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Example:
Recommendation: Do not use the incremental option for statistics gathering on these
objects.
Example:
--
Turn off the incremental option for 'SH.SALES':
dbms_stats.set_table_prefs('SH', 'SALES', 'INCREMENTAL', 'FALSE');
Rationale: The overhead of using the incremental option on these tables
outweighs the benefit of using the incremental option.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about REPORT_ADVISOR_TASK
18-31
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Prerequisites
To use DBMS_STATS.IMPLEMENT_ADVISOR_TASK, you must meet the following
prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
Note:
This subprogram executes using invoker's rights.
The results of performing this task depend on the privileges of the executing user:
• SYSTEM level
Only users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges
can perform this task on system-level rules.
• Operation level
The results depend on the following privileges:
– Users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges can
perform this task for all statistics operations.
18-32
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
– Users with the ANALYZE ANY privilege but not the ANALYZE ANY DICTIONARY
privilege can perform this task for statistics operations related to any schema
except SYS.
– Users with the ANALYZE ANY DICTIONARY privilege but not the ANALYZE ANY
privilege can perform this task for statistics operations related to their own
schema and the SYS schema.
– Users with neither the ANALYZE ANY nor the ANALYZE ANY DICTIONARY privilege
can only perform this operation for statistics operations relating to their own
schema.
• Object level
Users can perform this task for any object for which they have statistics collection
privileges.
For example, to implement all recommendations for the task opt_adv_task1, use
the following code:
IMP_RESULTS
------------------------------------
<implementation_results>
18-33
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
<rule NAME="AVOIDSETPROCEDURES">
<implemented>yes</implemented>
</rule>
<rule NAME="USEGATHERSCHEMASTATS">
<implemented>yes</implemented>
</rule>
<rule NAME="AVOIDSETPROCEDURES">
<implemented>yes</implemented>
</rule>
<rule NAME="USEGATHERSCHEMASTATS">
<implemented>yes</implemented>
</rule>
<rule NAME="USEDEFAULTPARAMS">
<implemented>no</implemented>
</rule>
<rule NAME="USEDEFAULTPARAMS">
<implemented>yes</implemented>
</rule>
<rule NAME="NOTUSEINCREMENTAL">
<implemented>yes</implemented>
</rule>
</implementation_results>
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.IMPLEMENT_ADVISOR_TASK
Prerequisites
To use the DBMS_STATS.SCRIPT_ADVISOR_TASK function, you must meet the following
prerequisites:
• To execute this subprogram, you must have the ADVISOR privilege.
• You must be the owner of the task.
18-34
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
Note:
This subprogram executes using invoker's rights.
The results of performing this task depend on the privileges of the executing user:
• SYSTEM level
Only users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges
can perform this task on system-level rules.
• Operation level
The results depend on the following privileges:
– Users with both the ANALYZE ANY and ANALYZE ANY DICTIONARY privileges can
perform this task for all statistics operations.
– Users with the ANALYZE ANY privilege but not the ANALYZE ANY DICTIONARY
privilege can perform this task for statistics operations related to any schema
except SYS.
– Users with the ANALYZE ANY DICTIONARY privilege but not the ANALYZE ANY
privilege can perform this task for statistics operations related to their own
schema and the SYS schema.
– Users with neither the ANALYZE ANY nor the ANALYZE ANY DICTIONARY privilege
can only perform this operation for statistics operations relating to their own
schema.
• Object level
Users can perform this task for any object for which they have statistics collection
privileges.
18-35
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
For example, to generate a script that contains recommendations for the task
opt_adv_task1, use the following code:
Note:
If you do not specify an execution name, then Optimizer Statistics
Advisor uses the most recent execution.
DECLARE
v_len NUMBER(10);
v_offset NUMBER(10) :=1;
v_amount NUMBER(10) :=10000;
BEGIN
v_len := DBMS_LOB.getlength(:b_report);
WHILE (v_offset < v_len)
LOOP
DBMS_OUTPUT.PUT_LINE(DBMS_LOB.SUBSTR(:b_script,v_amount,v_offset));
v_offset := v_offset + v_amount;
END LOOP;
END;
/
-- No scripts will be provided for the rule AVOIDSETPROCEDURES. Please check the
report
-- for more details.
-- No scripts will be provided for the rule USEGATHERSCHEMASTATS. Please check the
report
-- for more details.
-- No scripts will be provided for the rule AVOIDINEFFICIENTSTATSOPRSEQ. Please check
the
-- report for more details.
-- No scripts will be provided for the rule AVOIDUNNECESSARYSTATSCOLLECTION. Please
18-36
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
check
-- the report for more details.
-- No scripts will be provided for the rule GATHERSTATSAFTERBULKDML. Please check the
-- report for more details.
-- No scripts will be provided for the rule AVOIDDROPRECREATE. Please check the report
-- for more details.
-- No scripts will be provided for the rule AVOIDOUTOFRANGE. Please check the report
-- for more details.
-- No scripts will be provided for the rule AVOIDANALYZETABLE. Please check the report
-- for more details.
-- No scripts will be provided for the rule AVOIDSETPROCEDURES. Please check the
report
-- for more details.
-- No scripts will be provided for the rule USEGATHERSCHEMASTATS. Please check the
report
-- for more details.
-- No scripts will be provided for the rule AVOIDINEFFICIENTSTATSOPRSEQ. Please check
the
-- report for more details.
-- No scripts will be provided for the rule AVOIDUNNECESSARYSTATSCOLLECTION. Please
check
-- the report for more details.
-- No scripts will be provided for the rule GATHERSTATSAFTERBULKDML. Please check the
-- report for more details.
-- No scripts will be provided for the rule AVOIDDROPRECREATE. Please check the report
-- for more details.
-- No scripts will be provided for the rule AVOIDOUTOFRANGE. Please check the report
-- for more details.
-- No scripts will be provided for the rule AVOIDANALYZETABLE. Please check the report
-- for more details.
18-37
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
18-38
Chapter 18
Basic Tasks for Optimizer Statistics Advisor
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_STATS.SCRIPT_ADVISOR_TASK
18-39
Part VI
Optimizer Controls
You can use hints and initialization parameter to influence optimizer decisions and
behavior.
This part contains the following chapters:
• Influencing the Optimizer
Optimizer defaults are adequate for most operations, but not all.
• Improving Real-World Performance Through Cursor Sharing
Cursor sharing can improve database application performance by orders of
magnitude.
19
Influencing the Optimizer
Optimizer defaults are adequate for most operations, but not all.
In some cases you may have information unknown to the optimizer, or need to tune
the optimizer for a specific type of statement or workload. In such cases, influencing
the optimizer may provide better performance.
This chapter contains the following topics:
• Techniques for Influencing the Optimizer
You can influence the optimizer using several techniques, including SQL profiles,
SQL Plan Management, initialization parameters, and hints.
• Influencing the Optimizer with Initialization Parameters
This chapter explains which initialization parameters affect optimization, and how
to set them.
• Influencing the Optimizer with Hints
Optimizer hints are special comments in a SQL statement that pass instructions to
the optimizer.
19-1
Chapter 19
Techniques for Influencing the Optimizer
User
Optimizer
The overlapping squares in the preceding diagram show that SQL plan management
uses both initialization parameters and hints. SQL profiles also technically include
hints.
Note:
A stored outline is a legacy technique that serve a similar purpose to SQL
plan baselines.
19-2
Chapter 19
Influencing the Optimizer with Initialization Parameters
In some cases, multiple techniques optimize the same behavior. For example, you can
set optimizer goals using both initialization parameters and hints.
See Also:
"Migrating Stored Outlines to SQL Plan Baselines " to learn how to migrate
stored outlines to SQL plan baselines
19-3
Chapter 19
Influencing the Optimizer with Initialization Parameters
19-4
Chapter 19
Influencing the Optimizer with Initialization Parameters
19-5
Chapter 19
Influencing the Optimizer with Initialization Parameters
19-6
Chapter 19
Influencing the Optimizer with Initialization Parameters
19-7
Chapter 19
Influencing the Optimizer with Initialization Parameters
See Also:
For backward compatibility, you may not want the execution plans to change because
of new optimizer features in a new release. In such cases, you can set
OPTIMIZER_FEATURES_ENABLE to an earlier version. If you upgrade to a new release,
and if you want to enable the features in the new release, then you do not need to
explicitly set the OPTIMIZER_FEATURES_ENABLE initialization parameter.
Caution:
Oracle does not recommend explicitly setting the
OPTIMIZER_FEATURES_ENABLE initialization parameter to an earlier release. To
avoid SQL performance regression that may result from execution plan
changes, consider using SQL plan management instead.
Assumptions
This tutorial assumes the following:
• You recently upgraded the database from Oracle Database 12c Release 1 (12
1.0.2) to Oracle Database 12c Release 2 (12.2.0.1).
• You want to preserve the optimizer behavior from the earlier release.
19-8
Chapter 19
Influencing the Optimizer with Initialization Parameters
The preceding statement restores the optimizer functionality that existed in Oracle
Database 12c Release 1 (12.1.0.2).
See Also:
19-9
Chapter 19
Influencing the Optimizer with Initialization Parameters
Assumptions
This tutorial assumes the following:
• The primary application is interactive, so you want to set the optimizer goal for the
database instance to minimize response time.
• For the current session only, you want to run a report and optimize for throughput.
3. At the session level only, optimize for throughput before running a report.
For example, run the following SQL statement to configure only this session to
optimize for throughput:
See Also:
Oracle Database Reference to learn about the OPTIMIZER_MODE initialization
parameter
19-10
Chapter 19
Influencing the Optimizer with Initialization Parameters
Adaptive statistics are enabled when the following initialization parameters are set:
• OPTIMIZER_ADAPTIVE_STATISTICS is TRUE (the default is FALSE)
• OPTIMIZER_FEATURES_ENABLE is 12.1.0.1 or later
Assumptions
This tutorial assumes the following:
• The OPTIMIZER_FEATURES_ENABLE initialization parameter is set to 12.1.0.1 or
later.
• The OPTIMIZER_ADAPTIVE_REPORTING_ONLY initialization parameter is set to false
(default).
• You want to disable adaptive plans for testing purposes so that the database
generates only reports.
3. Run a query.
4. Run DBMS_XPLAN.DISPLAY_CURSOR with the +REPORT parameter.
When the +REPORT parameter is set, the report shows the plan the optimizer would
have picked if automatic reoptimization had been enabled.
19-11
Chapter 19
Influencing the Optimizer with Hints
See Also:
Note:
Oracle Database SQL Language Reference contains a complete reference
for all SQL hints
19-12
Chapter 19
Influencing the Optimizer with Hints
Optimizer
Generate Plan
Id Operation Name
0 SELECT STATEMENT
1 TABLE ACCESS BY INDEX ROWID EMPLOYEES
*2 INDEX UNIQUE SCAN EMP_DEP_IX
The advantage of hints is that they enable you to make decisions normally made by
the optimizer. In a test environment, hints are useful for testing the performance of a
specific access path. For example, you may know that an index is more selective for
certain queries, as in Figure 19-2. In this case, the hint may cause the optimizer to
generate a better plan.
The disadvantage of hints is the extra code that you must manage, check, and control.
Hints were introduced in Oracle7, when users had little recourse if the optimizer
generated suboptimal plans. Because changes in the database and host environment
can make hints obsolete or have negative consequences, a good practice is to test
using hints, but use other techniques to manage execution plans.
Oracle provides several tools, including SQL Tuning Advisor, SQL plan management,
and SQL Performance Analyzer, to address performance problems not solved by the
optimizer. Oracle strongly recommends that you use these tools instead of hints
because they provide fresh solutions as the data and database environment change.
This section contains the following topics:
• Types of Hints
You can use hints for tables, query blocks, and statements.
• Scope of Hints
When you specify a hint in a statement block, the hint applies to the appropriate
query block, table, or entire statement in the statement block. The hint overrides
any instance-level or session-level parameters.
• Guidelines for Hints
You must enclose hints within a SQL comment.
See Also:
Oracle Database SQL Language Reference for the most common hints by
functional category.
19-13
Chapter 19
Influencing the Optimizer with Hints
• Multitable
Multitable hints are like single-table hints except that the hint can specify multiple
tables or views. LEADING is an example of a multitable hint. The following
statement uses a multitable hint:
Note:
USE_NL(table1 table2) is not considered a multitable hint because it is
a shortcut for USE_NL(table1) and USE_NL(table2).
• Query block
Query block hints operate on single query blocks. STAR_TRANSFORMATION and
UNNEST are examples of query block hints. The following statement uses a query
block hint to specify that the FULL hint applies only to the query block that
references employees:
• Statement
Statement hints apply to the entire SQL statement. ALL_ROWS is an example of a
statement hint. The following statement uses a statement hint:
19-14
Chapter 19
Influencing the Optimizer with Hints
See Also:
Oracle Database SQL Language Reference for the most common hints by
functional category.
The preceding statement has two blocks, one for each component query. Hints in the
first component query apply only to its optimization, not to the optimization of the
second component query. For example, in the first week of 2015 you query current
year and last year sales. You apply FIRST_ROWS(10) to the query of last year's (2014)
sales and the ALL_ROWS hint to the query of this year's (2015) sales.
See Also:
Oracle Database SQL Language Reference for an overview of hints
The database ignores incorrectly specified hints. The database also ignores
combinations of conflicting hints, even if these hints are correctly specified. If one hint
19-15
Chapter 19
Influencing the Optimizer with Hints
is incorrectly specified, but a hint in the same comment is correctly specified, then the
database considers the correct hint.
Caution:
The database does not issue error messages for hints that it ignores.
A statement block can have only one comment containing hints, but it can contain
many space-separated hints. For example, a complex query may include multiple table
joins. If you specify only the INDEX hint for a specified table, then the optimizer must
determine the remaining access paths and corresponding join methods. The optimizer
may not use the INDEX hint because the join methods and access paths prevent it.
Example 19-2 uses multiple hints to specify the exact join order.
Example 19-2 Multiple Hints
See Also:
Oracle Database SQL Language Reference to learn about the syntax rules
for comments and hints
19-16
Chapter 19
Influencing the Optimizer with Hints
SELECT *
FROM taba a,
tabb b,
tabc c
WHERE a.acol BETWEEN 100 AND 200
AND b.bcol BETWEEN 10000 AND 20000
AND c.ccol BETWEEN 10000 AND 20000
AND a.key1 = b.key1
AND a.key2 = c.key2;
1. Choose the driving table and the driving index (if any).
Each of the first three conditions in the previous example is a filter condition that
applies to a single table. The last two conditions are join conditions.
Filter conditions dominate the choice of driving table and index. In general, the
driving table contains the filter condition that eliminates the highest percentage of
rows. Because the range of 100 to 200 is narrow compared with the range of acol,
but the ranges of 10000 and 20000 are relatively large, taba is the driving table, all
else being equal.
With nested loops joins, the joins occur through the join indexes, which are the
indexes on the primary or foreign keys used to connect that table to an earlier
table in the join tree. Rarely do you use the indexes on the non-join conditions,
except for the driving table. Thus, after taba is chosen as the driving table, use the
indexes on b.key1 and c.key2 to drive into tabb and tabc, respectively.
2. Choose the best join order, driving to the best unused filters earliest.
You can reduce the work of the following join by first joining to the table with the
best still-unused filter. Therefore, if bcol BETWEEN is more restrictive (rejects a
higher percentage of the rows) than ccol BETWEEN, then the last join becomes
easier (with fewer rows) if tabb is joined before tabc.
3. You can use the ORDERED or STAR hint to force the join order.
See Also:
Oracle Database Reference to learn about OPTIMIZER_MODE
19-17
20
Improving Real-World Performance
Through Cursor Sharing
Cursor sharing can improve database application performance by orders of
magnitude.
This chapter contains the following topics:
• Overview of Cursor Sharing
Oracle Database can share cursors, which are pointers to private SQL areas in the
shared pool.
• CURSOR_SHARING and Bind Variable Substitution
This topic explains what the CURSOR_SHARING initialization parameter is, and how
setting it to different values affects how Oracle Database uses bind variables.
• Adaptive Cursor Sharing
The adaptive cursor sharing feature enables a single statement that contains
bind variables to use multiple execution plans.
• Real-World Performance Guidelines for Cursor Sharing
The Real-World Performance team has created guidelines for how to optimize
cursor sharing in Oracle database applications.
20-1
Chapter 20
Overview of Cursor Sharing
When a server process executes SQL or PL/SQL code, the process uses the private
SQL area to store bind variable values, query execution state information, and query
execution work areas. The private SQL areas for each execution of a statement are
not shared and may contain different values and data.
A cursor is a name or handle to a specific private SQL area. The cursor contains
session-specific state information such as bind variable values and result sets.
As shown in the following graphic, you can think of a cursor as a pointer on the client
side and as a state on the server side. Because cursors are closely associated with
private SQL areas, the terms are sometimes used interchangeably.
PGA
SQL Work Areas Server
Process
Session Memory Private SQL Area
Cursor
Data Area
Client
Pointer Process
20-2
Chapter 20
Overview of Cursor Sharing
Instance
Shared Pool
Library Cache
Shared SQL Area Private
SELECT * FROM SQL Area
employees (Shared
Server Only)
PGA PGA
Server SQL Work Areas Server SQL Work Areas
Process Process
Session Memory Private SQL Area Session Memory Private SQL Area
Client Client
Process Process
20-3
Chapter 20
Overview of Cursor Sharing
Usually, SQL statements that differ only in literals cannot use the same shared
SQL area. For example, the following statements do not resolve to the same
SQL area:
The only exception to this rule is when the parameter CURSOR_SHARING has
been set to FORCE, in which case similar statements can share SQL areas.
See Also:
20-4
Chapter 20
Overview of Cursor Sharing
table in a different schema. Because of the syntactic identity, the third statement can
share a parent cursor with the first statement.
COUNT(*)
----------
319
COUNT(*)
----------
319
COUNT(*)
----------
155500
The following query of V$SQL indicates the two parents. The statement with the SQL ID
of 8h916vv2yw400, which is the lowercase “c” version of the statement, has one parent
cursor and two child cursors: child 0 and child 1. The statement with the SQL ID of
5rn2uxjtpz0wd, which is the uppercase “c” version of the statement, has a different
parent cursor and only one child cursor: child 0.
20-5
Chapter 20
Overview of Cursor Sharing
In the following example, a query of V$SQLAREA shows two parent cursors, each
identified with a different SQL_ID. The VERSION_COUNT indicates the number of child
cursors.
20-6
Chapter 20
Overview of Cursor Sharing
the following statement is not identical because the statement references different
employees tables for each user:
In the preceding results, the CHILD# of the bottom two statements is different (0 and 1),
even though the SQL_ID is the same. This means that the statements have the same
parent cursor, but different child cursors. In contrast, the statement with the SQL_ID of
5bzhzpaa0wy9m has one parent and one child (CHILD# of 0). All three SQL statements
use the same execution plan, as indicated by identical values in the PLAN_HASH_VALUE
column.
20-7
Chapter 20
Overview of Cursor Sharing
Video:
Video
20-8
Chapter 20
Overview of Cursor Sharing
the hard parse count to 49, but the second execution does not change the hard parse
count, which means that Oracle Database reused application code.
System altered.
NAME VALUE
------------------ ----------
parse count (hard) 48
COUNT(*)
----------
0
NAME VALUE
------------------ ----------
parse count (hard) 49
COUNT(*)
----------
0
NAME VALUE
------------------ ----------
parse count (hard) 49
20-9
Chapter 20
Overview of Cursor Sharing
SELECT value
FROM v$diag_info
WHERE name = 'Default Trace File';
VALUE
---------------------------------------------------------------------------
-----
/disk1/oracle/log/diag/rdbms/orcl/orcl/trace/orcl_ora_23054.trc
You enable tracing, use the TRACEFILE_IDENTIFIER initialization parameter to give the
trace file a meaningful name, and then query hr.employees:
Search the default trace file directory for the trace file that you generated:
% ls *emp_stmt.trc
orcl_ora_17950_emp_stmt.trc
Use TKPROF to format the trace file, and then open the formatted file:
The formatted trace file contains the parse information for the query of hr.employees.
SELECT *
FROM
hr.employees
20-10
Chapter 20
Overview of Cursor Sharing
A library cache miss indicates a hard parse. Performing the same steps for a second
execution of the same statement produces the following trace output, which shows no
library cache misses:
SELECT *
FROM
hr.employees
20-11
Chapter 20
Overview of Cursor Sharing
pw=0 time=961
us starts=1 cost=2 size=7383 card=107)
See Also:
"Shared Pool Check"
20-12
Chapter 20
Overview of Cursor Sharing
to latch the shared pool. This situation decreases concurrency and increases
contention.
Video:
Video
The following query of V$SQLAREA shows that the three statements require three
different parent cursors. As shown by VERSION_COUNT, each parent cursor requires its
own child cursor.
See Also:
"Do Not Use CURSOR_SHARING = FORCE as a Permanent Fix" to learn
about SQL injection
20-13
Chapter 20
Overview of Cursor Sharing
The Real-World Performance group has found that applications that use bind variables
perform better, scale better, and are more secure. Major benefits that result from using
bind variables include the following:
• Applications that use bind variables are not vulnerable to the same SQL injection
attacks as applications that use literals.
• When identical statements use bind variables, Oracle Database can take
advantage of cursor sharing, and share the plan and other information when
different values are bound to the same statement.
• Oracle Database avoids the overhead of latching the shared pool and library
cache required for hard parsing.
Video:
Video
The VERSION_COUNT value of 1 indicates that the database reused the same child
cursor rather than creating three separate child cursors. Using a bind variable made
this reuse possible.
20-14
Chapter 20
Overview of Cursor Sharing
The database hard parsed all three statements, which were not identical. The
DISPLAY_CURSOR output, which has been edited for clarity, shows that the optimizer
chose the same index range scan plan for the first two statements, but a full table scan
plan for the statement using literal 165:
---------------------------------------------------------------------------
----------
|Id| Operation | Name |Rows|Bytes|
Cost(%CPU)|Time|
---------------------------------------------------------------------------
----------
| 0| SELECT STATEMENT | | | |2
(100)| |
20-15
Chapter 20
Overview of Cursor Sharing
| 1| SORT AGGREGATE | |1 | 8 |
| |
| 2| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1 | 8 |2 (0) |
00:00:01 |
|*3| INDEX RANGE SCAN | EMP_EMP_ID_PK |1 | |1 (0) |
00:00:01 |
---------------------------------------------------------------------------
----------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("EMPLOYEE_ID"<101)
---------------------------------------------------------------------------
----------
|Id| Operation | Name |Rows|Bytes|
Cost(%CPU)|Time|
---------------------------------------------------------------------------
----------
| 0| SELECT STATEMENT | | | |2
(100)| |
| 1| SORT AGGREGATE | |1 | 8 |
| |
| 2| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |20|160|2 (0) |
00:00:01 |
|*3| INDEX RANGE SCAN | EMP_EMP_ID_PK |20| |1 (0) |
00:00:01 |
---------------------------------------------------------------------------
----------
3 - access("EMPLOYEE_ID"<120)
-------------------------------------------------------------------------
| Id | Operation | Name |Rows| Bytes |Cost(%CPU)| Time |
-------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 2 (100)| |
| 1 | SORT AGGREGATE | | 1 | 8 | | |
|* 2 | TABLE ACCESS FULL| EMPLOYEES | 66 | 528 | 2 (0)| 00:00:01 |
-------------------------------------------------------------------------
20-16
Chapter 20
Overview of Cursor Sharing
2 - filter("EMPLOYEE_ID"<165)
The preceding output shows that the optimizer considers a full table scan more
efficient than an index scan for the query that returns more rows.
Example 20-9 Bind Variables Result in Cursor Reuse
This example rewrites the queries executed in Example 20-8 to use bind variables
instead of literals. You bind the same values (101, 120, and 165) to the bind
variable :emp_id, and then display the execution plans for each:
The DISPLAY_CURSOR output shows that the optimizer chose exactly the same plan for
all three statements:
---------------------------------------------------------------------------
----------
| Id | Operation | Name |Rows|Bytes|Cost
(%CPU)|Time|
---------------------------------------------------------------------------
----------
| 0 | SELECT STATEMENT | | | |2
(100)| |
| 1 | SORT AGGREGATE | |1|8 |
| |
| 2 | TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1|8 | 2 (0)|
00:00:01 |
|* 3 | INDEX RANGE SCAN | EMP_EMP_ID_PK |1| | 1 (0)|
00:00:01 |
---------------------------------------------------------------------------
----------
3 - access("EMPLOYEE_ID"<:EMP_ID)
20-17
Chapter 20
Overview of Cursor Sharing
In contrast, when the preceding statements were executed with literals, the optimizer
chose a lower-cost full table scan when the employee ID value was 165. This is the
problem solved by adaptive cursor sharing.
See Also:
"Adaptive Cursor Sharing"
See Also:
Oracle Database Concepts for an overview of memory allocation in the
shared pool
20-18
Chapter 20
Overview of Cursor Sharing
• DBMS_STATS gathers statistics for a table, table cluster, or index when the
NO_INVALIDATE parameter is FALSE.
• A SQL statement references a schema object, which is later modified by a DDL
statement that uses immediate cursor invalidation (default).
You can manually specify immediate invalidation on statements such as ALTER
TABLE ... IMMEDIATE VALIDATION and ALTER INDEX ... IMMEDIATE VALIDATION,
or set the CURSOR_INVALIDATION initialization parameter to IMMEDIATE at the
session or system level.
Note:
A DDL statement using the DEFERRED VALIDATION clause overrides the
IMMEDIATE setting of the CURSOR_INVALIDATION initialization parameter.
When the preceding conditions are met, the database reparses the affected
statements at next execution.
When the database invalidates a cursor, the V$SQL.INVALIDATIONS value increases
(for example, from 0 to 1), and V$SQL.OBJECT_STATUS shows INVALID_UNAUTH.
COUNT(*)
----------
918843
SQL_ID
-------------
1y17j786c7jbh
20-19
Chapter 20
Overview of Cursor Sharing
See Also:
Note:
When V$SQL.IS_ROLLING_REFRESH_INVALID is Y, the underlying object has
changed, but recompilation of the cursor is not required. The database
updates metadata in the cursor.
20-20
Chapter 20
Overview of Cursor Sharing
Note:
If parallel SQL statements are marked rolling invalid, then the database
performs a hard parse at next execution, regardless of whether the time
period has expired. In an Oracle Real Application Clusters (Oracle RAC)
environment, this technique ensures consistency between execution plans of
parallel execution servers and the query coordinator.
An analogy for rolling invalidation might be the gradual replacement of worn-out office
furniture. Instead of replacing all the furniture at once, forcing a substantial financial
outlay, a company assigns each piece a different expiration date. Over the course of a
year, a piece stays in use until it is replaced, at which point a cost is incurred.
20-21
Chapter 20
CURSOR_SHARING and Bind Variable Substitution
See Also:
20-22
Chapter 20
CURSOR_SHARING and Bind Variable Substitution
Note:
The SIMILAR value for CURSOR_SHARING is deprecated.
You can set CURSOR_SHARING at the system or session level, or use the
CURSOR_SHARING_EXACT hint at the statement level.
See Also:
"Do Not Use CURSOR_SHARING = FORCE as a Permanent Fix"
Note:
If a statement uses an ORDER BY clause, then the database does not perform
literal replacement in the clause because it is not semantically correct to
consider the constant column number as a literal. The column number in the
ORDER BY clause affects the query plan and execution, so the database
cannot share two cursors having different column numbers.
When CURSOR_SHARING is set to FORCE, the database performs the following steps
during the parse:
1. Copies all literals in the statement to the PGA, and replaces them with system-
generated bind variables
For example, an application could process the following statement:
The optimizer replaces literals, including the literals in the SUBSTR function, as
follows:
2. Searches for an identical statement (same SQL hash value) in the shared pool
20-23
Chapter 20
CURSOR_SHARING and Bind Variable Substitution
If an identical statement is not found, then the database performs a hard parse.
Otherwise, the database proceeds to the next step.
3. Performs a soft parse of the statement
As the preceding steps indicate, setting the CURSOR_SHARING initialization parameter to
FORCE does not reduce the parse count. Rather, in some cases, FORCE enables the
database to perform a soft parse instead of a hard parse. Also, FORCE does not the
prevent against SQL injection attacks because Oracle Database binds the values after
any injection has already occurred.
Example 20-11 Replacement of Literals with System Bind Variables
This example sets CURSOR_SHARING to FORCE at the session level, executes three
statements containing literals, and displays the plan for each statement:
The following DISPLAY_CURSOR output, edited for readability, shows that all three
statements used the same plan. The optimizer chose the plan, an index range scan,
because it peeked at the first value (101) bound to the system bind variable, and
picked this plan as the best for all values. In fact, this plan is not the best plan for all
values. When the value is 165, a full table scan is more efficient.
---------------------------------------------------------------------------
----------
| Id | Operation | Name |Rows|Bytes|
Cost(%CPU)|Time|
---------------------------------------------------------------------------
----------
| 0 | SELECT STATEMENT | | | |2
(100)| |
| 1 | SORT AGGREGATE | |1 | 8 |
| |
| 2 | TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1 | 8 |2 (0)
|00:00:01|
|* 3 | INDEX RANGE SCAN | EMP_EMP_ID_PK |1 | |1 (0)
|00:00:01|
---------------------------------------------------------------------------
----------
Predicate Information (identified by operation id):
---------------------------------------------------
20-24
Chapter 20
Adaptive Cursor Sharing
3 - access("EMPLOYEE_ID"<101)
A query of V$SQLAREA confirms that Oracle Database replaced with the literal with
system bind variable :”SYS_B_0”, and created one parent and one child cursor
(VERSION_COUNT=1) for all three statements, which means that all executions shared
the same plan.
See Also:
• "Private and Shared SQL Areas" for more details on the various checks
performed
• Oracle Database Reference to learn about the CURSOR_SHARING
initialization parameter
20-25
Chapter 20
Adaptive Cursor Sharing
• Cursor Merging
If the optimizer creates a plan for a bind-aware cursor, and if this plan is the same
as an existing cursor, then the optimizer can perform cursor merging.
• Adaptive Cursor Sharing Views
You can use the V$ views for adaptive cursor sharing to see selectivity ranges,
cursor information (such as whether a cursor is bind-aware or bind-sensitive), and
execution statistics.
Note:
Adaptive cursor sharing is independent of the CURSOR_SHARING initialization
parameter. Adaptive cursor sharing is equally applicable to statements that
contain user-defined and system-generated bind variables. Adaptive cursor
sharing does not apply to statements that contain only literals.
Also assume in this example that a histogram exists on at least one of the columns in
the predicate. The database processes this statement as follows:
1. The application issues the statement for the first time, which causes a hard parse.
During the parse, the database performs the following tasks:
• Peeks at the bind variables to generate the initial plan.
20-26
Chapter 20
Adaptive Cursor Sharing
11. The database stores the new bind values and execution statistics in the new child
cursor.
20-27
Chapter 20
Adaptive Cursor Sharing
12. The application issues the statement a fifth time, using different bind variables,
which causes a soft parse. Because the cursor is bind-aware, the database does
the following:
• Determines whether the cardinality of the new values falls within the same
range as the stored cardinality. In this example, the cardinality is 20.
• Does not find a matching child cursor.
13. The database performs a hard parse. As a result, the database does the following:
a. Creates a new child cursor with a third execution plan (in this example, index
access)
b. Determines that this index access execution plan is the same as the index
access execution plan used for the first execution of the statement
c. Merges the two child cursors containing index access plans, which involves
storing the combined cardinality statistics into one child cursor, and deleting
the other one
14. The database executes the cursor using the index access execution plan.
DECLARE
v_counter NUMBER(7) := 1000;
BEGIN
FOR i IN 1..100000 LOOP
INSERT INTO hr.employees
20-28
Chapter 20
Adaptive Cursor Sharing
VALUES (v_counter,null,'Doe','Doe@example.com',null,'07-
JUN-02','AC_ACCOUNT',null,null,null,50);
v_counter := v_counter + 1;
END LOOP;
END;
/
COMMIT;
COUNT(*) MAX(EMPLOYEE_ID)
---------- ----------------
1 200
The optimizer chooses an index range scan, as expected for such a low-cardinality
query:
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID a9upgaqqj7bn5, child number 0
-------------------------------------
20-29
Chapter 20
Adaptive Cursor Sharing
---------------------------------------------------------------------------
----------
| Id| Operation | Name |Rows|Bytes|Cost
(%CPU)|Time |
---------------------------------------------------------------------------
----------
| 0| SELECT STATEMENT | | | |
2(100)| |
| 1| SORT AGGREGATE | |1 |8 |
| |
| 2| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1 |8 |2
(0)|00:00:01|
|*3| INDEX RANGE SCAN | EMP_DEPARTMENT_IX |1 | |1
(0)|00:00:01|
---------------------------------------------------------------------------
----------
3 - access("DEPARTMENT_ID"=:DEPT_ID)
20-30
Chapter 20
Adaptive Cursor Sharing
The preceding output shows one child cursor that has been executed once for the low-
cardinality query. The cursor has been marked bind-sensitive because the optimizer
believes the optimal plan may depend on the value of the bind variable.
When a cursor is marked bind-sensitive, Oracle Database monitors the behavior of the
cursor using different bind values, to determine whether a different plan for different
bind values is more efficient. The database marked this cursor bind-sensitive because
the optimizer used the histogram on the department_id column to compute the
selectivity of the predicate WHERE department_id = :dept_id. Because the presence
of the histogram indicates that the column is skewed, different values of the bind
variable may require different plans.
Example 20-14 High-Cardinality Query
This example continues the example in Example 20-13. The following code re-
executes the same query using the value 50, which occupies 99.9% of the rows:
COUNT(*) MAX(EMPLOYEE_ID)
---------- ----------------
100045 100999
Even though such an unselective query would be more efficient with a full table scan,
the optimizer chooses the same index range scan used for department_id=10. This
reason is that the database assumes that the existing plan in the cursor can be
shared:
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID a9upgaqqj7bn5, child number 0
-------------------------------------
SELECT COUNT(*), MAX(employee_id) FROM hr.employees WHERE department_id
= :dept_id
---------------------------------------------------------------------------
----------
| Id| Operation | Name |Rows|Bytes|Cost
(%CPU)|Time |
---------------------------------------------------------------------------
----------
| 0| SELECT STATEMENT | | | |
2(100)| |
| 1| SORT AGGREGATE | |1 |8 |
| |
| 2| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1 |8 |2
(0)|00:00:01|
|*3| INDEX RANGE SCAN | EMP_DEPARTMENT_IX |1 | |1
(0)|00:00:01|
20-31
Chapter 20
Adaptive Cursor Sharing
---------------------------------------------------------------------------
----------
3 - access("DEPARTMENT_ID"=:DEPT_ID)
A query of V$SQL shows that the child cursor has now been executed twice:
At this stage, the optimizer has not yet marked the cursor as bind-aware.
See Also:
Oracle Database Reference to learn about V$SQL
20-32
Chapter 20
Adaptive Cursor Sharing
• Marks the original cursor generated for the statement as not sharable
(V$SQL.IS_SHAREABLE is N). The original cursor is no longer usable and is eligible
to age out of the library cache
When the same query repeatedly executes with different bind values, the database
adds new bind values to the “signature” of the SQL statement (which includes the
optimizer environment, NLS settings, and so on), and categorizes the values. The
database examines the bind values, and considers whether the current bind value
results in a significantly different data volume, or whether an existing plan is sufficient.
The database does not need to create a new plan for each new value.
Consider a scenario in which you execute a statement with 12 distinct bind values
(executing each distinct value twice), which causes the database to trigger 5 hard
parses, and create 2 additional plans. Because the database performs 5 hard parses,
it creates 5 new child cursors, even though some cursors have the same execution
plan as existing cursors. The database marks the superfluous cursors as not usable,
which means these cursors eventually age out of the library cache.
During the initial hard parses, the optimizer is essentially mapping out the relationship
between bind values and the appropriate execution plan. After this initial period, the
database eventually reaches a steady state. Executing with a new bind value results in
picking the best child cursor in the cache, without requiring a hard parse. Thus, the
number of parses does not scale with the number of different bind values.
Example 20-15 Bind-Aware Cursors
This example continues the example in "Bind-Sensitive Cursors". The following code
issues a second query employees with the bind variable set to 50:
COUNT(*) MAX(EMPLOYEE_ID)
---------- ----------------
100045 100999
During the first two executions, the database was monitoring the behavior of the
queries, and determined that the different bind values caused the queries to differ
significantly in cardinality. Based on this difference, the database adapts its behavior
so that the same plan is not always shared for this query. Thus, the optimizer
generates a new plan based on the current bind value, which is 50:
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID a9upgaqqj7bn5, child number 1
-------------------------------------
SELECT COUNT(*), MAX(employee_id) FROM hr.employees WHERE department_id
= :dept_id
---------------------------------------------------------------------------
--
20-33
Chapter 20
Adaptive Cursor Sharing
2 - filter("DEPARTMENT_ID"=:DEPT_ID)
The preceding output shows that the database created an additional child cursor
(CHILD# of 1). Cursor 0 is now marked as not shareable. Cursor 1 shows a number of
buffers gets lower than cursor 0, and is marked both bind-sensitive and bind-aware. A
bind-aware cursor may use different plans for different bind values, depending on the
selectivity of the predicates containing the bind variable.
20-34
Chapter 20
Adaptive Cursor Sharing
COUNT(*) MAX(EMPLOYEE_ID)
---------- ----------------
1 200
The following output shows that the optimizer picked the best plan, which is an index
scan, based on the low cardinality estimate for the current bind value of 10:
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID a9upgaqqj7bn5, child number 2
-------------------------------------
select COUNT(*), MAX(employee_id) FROM hr.employees WHERE department_id
= :dept_id
---------------------------------------------------------------------------
----------
| Id| Operation | Name |Rows|Bytes|Cost
(%CPU)|Time |
---------------------------------------------------------------------------
----------
| 0| SELECT STATEMENT | | | |
2(100)| |
| 1| SORT AGGREGATE | |1 |8 |
| |
| 2| TABLE ACCESS BY INDEX ROWID BATCHED| EMPLOYEES |1 |8 |2
(0)|00:00:01|
|*3| INDEX RANGE SCAN | EMP_DEPARTMENT_IX | 1| |1
(0)|00:00:01|
---------------------------------------------------------------------------
----------
3 - access("DEPARTMENT_ID"=:DEPT_ID)
The V$SQL output now shows that three child cursors exist:
20-35
Chapter 20
Adaptive Cursor Sharing
The database discarded the original cursor (CHILD# of 0) when the cursor switched to
bind-aware mode. This is a one-time overhead. The database marked cursor 0 as not
shareable (SHAREABLE is N), which means that this cursor is unusable and will be
among the first to age out of the cursor cache.
See Also:
Oracle Database Reference to learn about V$SQL
20-36
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
one cursor as not sharable. If the library cache is under space pressure, then the
database ages out the non-sharable cursor first.
See Also:
"Example 20-12"
See Also:
Oracle Database Reference to learn about V$SQL and its related views
20-37
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
Video:
Video
20-38
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
becomes. When multiple statements share the same execution plan, the requests
for latches and the durations of latches go down. This behavior increases
scalability.
• Throughput and response time
When the database avoids constantly reparsing and creating cursors, more of its
time is spent in user space. The Real-World Performance group has found that
changing literals to use binds often leads to orders of magnitude improvements in
throughput and user response time.
Video:
Video
• Security
The only way to prevent SQL injection attacks is to use bind variables. Malicious
users can exploit application that concatenate strings by “injecting” code into the
application.
See Also:
Oracle Database PL/SQL Language Reference for an example of an
application that fixes a security vulnerability created by literals
20-39
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
• It indicates that the application does not use user-defined bind variables, which
means that it is open to SQL injection. Setting CURSOR_SHARING to FORCE does not
fix SQL injection bugs or render the code any more secure. The database binds
values only after any malicious SQL text has already been injected.
Video:
Video
• The database must perform extra work during the soft parse to find a similar
statement in the shared pool.
• The database removes every literal, which means that it can remove useful
information. For example, the database strips out literal values in SUBSTR and
TO_DATE functions. The use of system-generated bind variables where literals are
more optimal can have a negative impact on execution plans.
• There is an increase in the maximum lengths (as returned by DESCRIBE) of any
selected expressions that contain literals in a SELECT statement. However, the
actual length of the data returned does not change.
• Star transformation is not supported.
See Also:
The Real-World Performance group recommends that you standardize spacing and
capitalization conventions for SQL statements and PL/SQL blocks. Also establish
conventions for the naming and definitions of bind variables. If the database does not
share cursors as expected, begin your diagnosis by querying V$SQL_SHARED_CURSOR.
20-40
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
7 rows selected.
The following query shows that the database did not share the cursor:
20-41
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
20-42
Chapter 20
Real-World Performance Guidelines for Cursor Sharing
20-43
Part VII
Monitoring and Tracing SQL
Use DBMS_MONITOR to track database operations, SQL Test Case Builder to package
information relating to a performance problem, and SQL Trace to generate diagnostic
data for problem SQL statements.
This part contains the following chapters:
• Monitoring Database Operations
This chapter describes how to monitor SQL and PL/SQL.
• Gathering Diagnostic Data with SQL Test Case Builder
SQL Test Case Builder is a tool that automatically gathers information needed to
reproduce the problem in a different database instance.
• Performing Application Tracing
This chapter explains what end-to-end application tracing is, and how to generate
and read trace files.
21
Monitoring Database Operations
This chapter describes how to monitor SQL and PL/SQL.
This chapter contains the following topics:
• About Monitoring Database Operations
A database operation is a user-defined logical object that includes session
activity between two points in time.
• Enabling and Disabling Monitoring of Database Operations
Use initialization parameters to enable or disable monitoring.
• Defining a Composite Database Operation
Defining a database operation involves supplying a name and specifying its
beginning and end times.
• Monitoring SQL Executions Using Cloud Control
By default, AWR automatically captures SQL monitoring reports in XML format.
21-1
Chapter 21
About Monitoring Database Operations
See Also:
Oracle Database Concepts for a brief conceptual overview of database
operations
21-2
Chapter 21
About Monitoring Database Operations
terminate the session or call, or log the event. You can then monitor these SQL
operations.
• Tuning for response time
When tuning a database operation, you typically aim to improve the response
time. Often the database operation performance issues are mainly SQL
performance issues.
This section contains the following topics:
See Also:
21-3
Chapter 21
About Monitoring Database Operations
• After a database upgrade, the execution time of an important batch job doubled.
To resolve this problem, you must collect enough relevant statistical data from the
batch job before and after the upgrade, compare the two sets of data, and then
identify the changes.
• Packing a SQL tuning set (STS) took far longer than anticipated. To diagnose the
problem, you need to know what was being executed over time. Because this
issue cannot be easily reproduced, you need to monitor the process while it is
running.
See Also:
"About SQL Tuning Sets"
The following figure gives an overview of the architecture for Real-Time SQL
Monitoring.
21-4
Chapter 21
About Monitoring Database Operations
DBMS_SQL_MONITOR
REPORT_SQL_MONITOR
Oracle
Enterprise
Manager
User
CONTROL_MANAGEMENT_PACK_ACCESS
21-5
Chapter 21
About Monitoring Database Operations
See Also:
21-6
Chapter 21
About Monitoring Database Operations
SQL statements or PL/SQL subprograms that execute within these two points in time
are part of the composite operation. A session can participate in at most one
composite database operation at a time.
Oracle Database monitors composite database operations when either of the following
conditions is true:
• You started an operation with DBMS_SQL_MONITOR.BEGIN_OPERATION, and the
operation has consumed at least 5 seconds of CPU or I/O time.
• Tracking for the operations is forced by setting FORCE_TRACKING to Y in
BEGIN_OPERATION.
In OLTP and data warehouse environments, a job often logically groups related SQL
statements. The job can span multiple concurrent sessions. Typical use cases for
monitoring composite operations include the following:
• A periodic batch job containing many SQL statements must complete in a certain
number of hours, but took longer than expected.
• After a database upgrade, the execution time of an important batch job increased.
To resolve this problem, you must collect enough relevant statistical data from the
batch job before and after the upgrade, compare the two sets of data, and then
identify the changes.
• Packing a SQL tuning set (STS) took far longer than anticipated. To diagnose the
problem, you need to know what was being executed over time. Because this
issue cannot be easily reproduced, you need to monitor the process while it is
running.
See Also:
21-7
Chapter 21
About Monitoring Database Operations
21-8
Chapter 21
About Monitoring Database Operations
Subprogram Description
BEGIN_OPERATIO This function starts a database operation in the current session.
N This function associates a session with a database operation. Starting in
Oracle Database 12c Release 2 (12.2), you can use session_id and
session_num to indicate the session in which to start monitoring.
END_OPERATION This function ends a database operation in the current session. If the
specified database operation does not exist, then this function has no
effect.
21-9
Chapter 21
About Monitoring Database Operations
Subprogram Description
REPORT_SQL_MON This function builds a detailed report with monitoring information for a SQL
ITOR statement, PL/SQL block, or database operation.
For each operation, it gives key information and associated global
statistics. Use this function to get detailed monitoring information for a
database operation.
The target database operation for this report can be:
• The last database operation monitored by Oracle Database (default,
no parameter).
• The last database operation executed in the specified session and
monitored by Oracle Database. The session is identified by its session
ID and optionally its serial number (-1 is current session).
• The last execution of a specific database operation identified by its
sql_id.
• A specific execution of a database operation identified by the
combination sql_id, sql_exec_start, and sql_exec_id.
• The last execution of a specific database operation identified by
dbop_name.
• The specific execution of a database operation identified by the
combination dbop_name, dbop_exec_id.
Use the type parameter to specify the output type: TEXT (default), HTML,
ACTIVE, or XML.
REPORT_SQL_MON This function is identical to the REPORT_SQL_MONITOR function, except
ITOR_XML that the return type is XMLType.
REPORT_SQL_MON This function builds a report for all or a subset of database operations that
ITOR_LIST have been monitored by Oracle Database.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_SQL_MONITOR package
21-10
Chapter 21
About Monitoring Database Operations
View Description
DBA_HIST_REPORTS This view displays metadata about XML reports captured into
Automatic Workload Repository (AWR). Each XML report
contains details about some activity of a component. For
example, a SQL Monitor report contains a detailed report
about a particular database operation.
Important columns include:
• The REPORT_SUMMARY column contains the summary of
the report.
• The COMPONENT_NAME column accepts the value
sqlmonitor.
• The REPORT_ID column provides the ID of the report,
which you can specify in the RID parameter of
DBMS_AUTO_REPORT.REPORT_REPOSITORY_DETAIL.
DBA_HIST_REPORTS_DETAILS This view displays details about each report captured in AWR.
Metadata for each report appears in the DBA_HIST_REPORTS
view, whereas the actual report is available in the
DBA_HIST_REPORTS_DETAILS view.
V$SQL_MONITOR This view contains global, high-level information about simple
and composite database operations.
For simple database operations, monitoring statistics are not
cumulative over several executions. In this case, one entry in
V$SQL_MONITOR is dedicated to a single execution of a SQL
statement. If the database monitors two executions of the
same SQL statement, then each execution has a separate
entry in V$SQL_MONITOR.
For simple database operations, V$SQL_MONITOR has one
entry for the parallel execution coordinator process and one
entry for each parallel execution server process. Each entry
has corresponding entries in V$SQL_PLAN_MONITOR.
Because the processes allocated for the parallel execution of
a SQL statement are cooperating for the same execution,
these entries share the same execution key (the combination
of SQL_ID, SQL_EXEC_START, and SQL_EXEC_ID).
For composite database operations, each row contains an
operation whose statistics are accumulated over the SQL
statements and PL/SQL subprograms that run in the same
session as part of the operation. The primary key is the
combination of the columns DBOP_NAME and DBOP_EXEC_ID.
V$SQL_MONITOR_SESSTAT This view contains the statistics for all sessions involved in the
database operation.
Most of the statistics are cumulative. The database stores the
statistics in XML format instead of using each column for each
statistic. This view is primarily intended for the report
generator. Oracle recommends that you use V$SESSTAT
instead of V$SQL_MONITOR_SESSTAT.
21-11
Chapter 21
About Monitoring Database Operations
View Description
V$SQL_PLAN_MONITOR This view contains monitoring statistics for each step in the
execution plan of the monitored SQL statement.
The database updates statistics in V$SQL_PLAN_MONITOR
every second while the SQL statement is executing. Multiple
entries exist in V$SQL_PLAN_MONITOR for every monitored
SQL statement. Each entry corresponds to a step in the
execution plan of the statement.
You can use the preceding V$ views with the following views to get additional
information about the monitored execution:
• V$ACTIVE_SESSION_HISTORY
• V$SESSION
• V$SESSION_LONGOPS
• V$SQL
• V$SQL_PLAN
See Also:
Oracle Database Reference to learn about the V$ views for database
operations monitoring
21-12
Chapter 21
Enabling and Disabling Monitoring of Database Operations
Prerequisites
Because SQL monitoring is a feature of the Oracle Database Tuning Pack, the
CONTROL_MANAGEMENT_PACK_ACCESS initialization parameter must be set to DIAGNOSTIC
+TUNING (the default value).
Assumptions
This tutorial assumes the following:
• The STATISTICS_LEVEL initialization parameter is set to BASIC.
• You want to enable automatic monitoring of database operations.
21-13
Chapter 21
Enabling and Disabling Monitoring of Database Operations
See Also:
Oracle Database Reference to learn about the STATISTICS_LEVEL and
CONTROL_MANAGEMENT_PACK_ACCESS initialization parameter
Assumptions
This tutorial assumes the following:
• Database monitoring is currently enabled at the system level.
• You want to disable automatic monitoring for the statement SELECT * FROM sales
ORDER BY time_id.
21-14
Chapter 21
Defining a Composite Database Operation
See Also:
Oracle Database SQL Language Reference for information about using the
MONITOR and NO_MONITOR hints
A single namespace exists for database operations, which means that name collisions
are possible. Oracle recommends the following naming convention:
component_name.subcomponent_name.operation name. For operations inside the
database, Oracle recommends using ORA for the component name. For example, a
materialized view refresh could be named ORA.MV.refresh. An E-Business Suite
payroll function could be named EBIZ.payroll.
3. Execute the SQL statements or PL/SQL programs that you want to monitor.
4. End the operation by using DBMS_SQL_MONITOR.END_OPERATION.
The following example ends operation ORA.sales.agg:
BEGIN
DBMS_SQL_MONITOR.END_OPERATION ( dbop_name => 'ORA.sales.agg',
dbop_eid => :exec_id );
END;
/
21-15
Chapter 21
Defining a Composite Database Operation
SID SERIAL#
---------- ----------
121 13397
PRINT eid
EID
----------
2
21-16
Chapter 21
Defining a Composite Database Operation
SELECT COUNT(*)
FROM sh.customers;
COUNT(*)
----------
55500
21-17
Chapter 21
Monitoring SQL Executions Using Cloud Control
Assumptions
This tutorial assumes the following:
21-18
Chapter 21
Monitoring SQL Executions Using Cloud Control
• The user sh is executing the following long-running parallel query of the sales
made to each customer:
• You want to ensure that the preceding query does not consume excessive
resources. While the statement executes, you want to determine basic statistics
about the database operation, such as the level of parallelism, the total database
time, and number of I/O requests.
• You use Cloud Control to monitor statement execution.
Note:
To generate the SQL monitor report from the command line, run the
REPORT_SQL_MONITOR function in the DBMS_SQLTUNE package, as in the
following sample SQL*Plus script:
In this example, the query has been executing for 1.4 minutes.
2. Click the value in the SQL ID column to see details about the statement.
The Monitored SQL Details page appears.
21-19
Chapter 21
Monitoring SQL Executions Using Cloud Control
The preceding report shows the execution plan and statistics relating to statement
execution. For example, the Timeline column shows when each step of the
execution plan was active. Times are shown relative to the beginning and end of
the statement execution. The Executions column shows how many times an
operation was executed.
3. In the Overview section, click the link next to the SQL text.
A message shows the full text of the SQL statement.
4. In the Time & Wait Statistics section, next to Database Time, move the cursor over
the largest portion on the bar graph.
A message shows that user I/O is consuming over half of database time.
21-20
Chapter 21
Monitoring SQL Executions Using Cloud Control
Database Time measures the amount of time the database has spent working on
this SQL statement. This value includes CPU and wait times, such as I/O time.
The bar graph is divided into several color-coded portions to highlight CPU
resources, user I/O resources, and other resources. You can move the cursor over
any portion to view the percentage value of the total.
5. In the Details section, in the IO Requests column, move the cursor over the I/O
requests bar to view the percentage value of the total.
A message appears.
In the preceding graphic, the IO Requests message shows the total number of
read requests issued by the monitored SQL. The message shows that read
requests form 80% of the total I/O requests.
See Also:
21-21
22
Gathering Diagnostic Data with SQL Test
Case Builder
SQL Test Case Builder is a tool that automatically gathers information needed to
reproduce the problem in a different database instance.
A SQL test case is a set of information that enables a developer to reproduce the
execution plan for a specific SQL statement that has encountered a performance
problem.
This chapter contains the following topics:
• Purpose of SQL Test Case Builder
SQL Test Case Builder automates the process of gathering and reproducing
information about a problem and the environment in which it occurred.
• Concepts for SQL Test Case Builder
Key concepts for SQL Test Case Builder include SQL incidents, types of
information recorded, and the form of the output.
• User Interfaces for SQL Test Case Builder
You can access SQL Test Case Builder either through Cloud Control or using
PL/SQL on the command line.
• Running SQL Test Case Builder
You can run SQL Test Case Builder using Cloud Control.
22-1
Chapter 22
Concepts for SQL Test Case Builder
• SQL Incidents
In the fault diagnosability infrastructure of Oracle Database, an incident is a single
occurrence of a problem.
• What SQL Test Case Builder Captures
SQL Test Case Builder captures permanent information about a SQL query and its
environment.
• Output of SQL Test Case Builder
The output of the SQL Test Case Builder is a set of files that contains commands
required to re-create the environment and all necessary objects.
See Also:
22-2
Chapter 22
Concepts for SQL Test Case Builder
The database automatically handles the memory requested for each SQL
operation. Actions such as sorting can affect performance significantly. SQL Test
Case Builder keeps track of the memory activities, for example, where the
database allocated memory and how much it allocated.
• Dynamic statistics
Regathering dynamic statistics on a different database does not always generate
the same results, for example, when data is missing. To reproduce the problem,
SQL Test Case Builder exports the dynamic statistics result from the source
database. In the testing database, SQL Test Case Builder reuses the same values
captured from the source database instead of regathering dynamic statistics.
• Multiple execution support
SQL Test Case Builder can capture dynamic information accumulated during
multiple executions of the query. This capability is important for automatic
reoptimization.
• Compilation environment and bind values replay
The compilation environment setting is an important part of the query optimization
context. SQL Test Case Builder captures nondefault settings altered by the user
when running the problem query in the source database. If any nondefault
parameter values are used, SQL Test Case Builder re-establishes the same
values before running the query.
• Object statistics history
The statistics history for objects is helpful to determine whether a plan change was
caused by a change in statistics values. DBMS_STATS stores the history in the data
dictionary. SQL Test Case Builder stores this statistics data into a staging table
during export. During import, SQL Test Case Builder automatically reloads the
statistics history data into the target database from the staging table.
• Statement history
The statement history is important for diagnosing problems related to adaptive
cursor sharing, statistics feedback, and cursor sharing bugs. The history includes
execution plans and compilation and execution statistics.
See Also:
22-3
Chapter 22
User Interfaces for SQL Test Case Builder
By default, SQL Test Case Builder stores the files in the following location, where
incnum refers to the incident number and runnum refers to the run number:
$ADR_HOME/incident/incdir_incnum/SQLTCB_runnum
$ORACLE_HOME/log/diag/rdbms/dbsa/dbsa/incident/incdir_2657/SQLTCB_1
You can specify a nondefault location by creating an Oracle directory and invoking
DBMS_SQLDIAG.EXPORT_SQL_TESTCASE, as in the following example:
BEGIN
DBMS_SQLDIAG.EXPORT_SQL_TESTCASE (
directory => 'my_tcb_dir_exp'
, sql_text => 'SELECT COUNT(*) FROM sales'
, testcase => tco
);
END;
See Also:
Oracle Database Administrator’s Guide to learn about the structure of the
ADR repository
22-4
Chapter 22
User Interfaces for SQL Test Case Builder
The Support Workbench page appears, with the incidents listed in a table.
22-5
Chapter 22
User Interfaces for SQL Test Case Builder
See Also:
See Also:
Online help for Cloud Control
This package consists of various subprograms for the SQL Test Case Builder, some of
which are listed in the following table.
Procedure Description
EXPORT_SQL_TESTCASE Exports a SQL test case to a user-specified directory
EXPORT_SQL_TESTCASE_DIR_BY_INC Exports a SQL test case corresponding to the incident
ID passed as an argument
EXPORT_SQL_TESTCASE_DIR_BY_TXT Exports a SQL test case corresponding to the SQL
text passed as an argument
IMPORT_SQL_TESTCASE Imports a SQL test case into a schema
22-6
Chapter 22
Running SQL Test Case Builder
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DBMS_SQLDIAG package
Assumptions
This tutorial assumes the following:
• You ran the following EXPLAIN PLAN statement as user sh, which causes an
internal error:
• In the Incidents and Problems section on the Database Home page, a SQL
incident generated by the internal error appears.
• You access the Incident Details page, as explained in "Accessing the Incident
Manager".
22-7
Chapter 22
Running SQL Test Case Builder
22-8
Chapter 22
Running SQL Test Case Builder
7. Access the SQL Test Case files in the location described in "Output of SQL Test
Case Builder".
See Also:
Online help for Cloud Control
22-9
23
Performing Application Tracing
This chapter explains what end-to-end application tracing is, and how to generate and
read trace files.
This chapter contains the following topics:
See Also:
SQL*Plus User's Guide and Reference to learn about the use of Autotrace to
trace and tune SQL*Plus statements
23-1
Chapter 23
Overview of End-to-End Application Tracing
23-2
Chapter 23
Overview of End-to-End Application Tracing
See Also:
"Views for Application Tracing"
23-3
Chapter 23
Overview of End-to-End Application Tracing
See Also:
Oracle Database PL/SQL Packages and Types Reference for information
about the DBMS_MONITOR, DBMS_SESSION, DBMS_SERVICE, and
DBMS_APPLICATION_INFO packages
23-4
Chapter 23
Enabling Statistics Gathering for End-to-End Tracing
See Also:
• "TRCSESS"
• "Enabling End-to-End Application Tracing" to learn how to use the
DBMS_SESSION or DBMS_MONITOR packages to enable SQL tracing for a
session or an instance
The default level is the session-level statistics gathering. Statistics gathering is global
for the database and continues after a database instance is restarted.
You can gather statistics by the following criteria:
• Enabling Statistics Gathering for a Client ID
The procedure CLIENT_ID_STAT_ENABLE enables statistics gathering for a given
client ID, whereas the procedure CLIENT_ID_STAT_DISABLE disables it.
• Enabling Statistics Gathering for Services, Modules, and Actions
The procedure SERV_MOD_ACT_STAT_ENABLE enables statistic gathering for a
combination of service, module, and action, whereas the procedure
SERV_MOD_ACT_STAT_DISABLE disables statistic gathering for a combination of
service, module, and action.
23-5
Chapter 23
Enabling Statistics Gathering for End-to-End Tracing
Assumptions
This tutorial assumes that you want to enable and then disable statistics gathering for
the client with the ID oe.oe.
Assumptions
This tutorial assumes that you want to gather statistics as follows:
• For the ACCTG service
• For all actions in the PAYROLL module
• For the INSERT ITEM action within the GLEDGER module
To enable and disable statistics gathering for a service, module, and action:
1. Start SQL*Plus, and then connect to the database with the appropriate privileges.
2. Enable statistics gathering for the desired service, module, and action.
For example, run the following commands:
BEGIN
DBMS_MONITOR.SERV_MOD_ACT_STAT_ENABLE(
service_name => 'ACCTG'
, module_name => 'PAYROLL' );
23-6
Chapter 23
Enabling End-to-End Application Tracing
END;
BEGIN
DBMS_MONITOR.SERV_MOD_ACT_STAT_ENABLE(
service_name => 'ACCTG'
, module_name => 'GLEDGER'
, action_name => 'INSERT ITEM' );
END;
BEGIN
DBMS_MONITOR.SERV_MOD_ACT_STAT_DISABLE(
service_name => 'ACCTG'
, module_name => 'GLEDGER'
, action_name => 'INSERT ITEM' );
END;
With the criteria that you provide, specific trace information is captured in a set of trace
files and combined into a single output trace file. You can enable tracing for specific
diagnosis and workload management by the following criteria:
See Also:
Oracle Database Administrator’s Guide for information about how to locate
trace files
23-7
Chapter 23
Enabling End-to-End Application Tracing
Assumptions
This tutorial assumes the following:
• OE.OE is the client identifier for which SQL tracing is to be enabled.
• You want to include wait information in the trace.
• You want to exclude bind information from the trace.
BEGIN
DBMS_MONITOR.CLIENT_ID_TRACE_ENABLE(
client_id => 'OE.OE' ,
waits => true ,
binds => false );
END;
Assumptions
This tutorial assumes the following:
• You want to enable tracing for the service ACCTG.
• You want to enable tracing for all actions for the combination of the ACCTG service
and the PAYROLL module.
• You want to include wait information in the trace.
23-8
Chapter 23
Enabling End-to-End Application Tracing
BEGIN
DBMS_MONITOR.SERV_MOD_ACT_TRACE_ENABLE(
service_name => 'ACCTG' ,
module_name => 'PAYROLL' ,
waits => true ,
binds => false ,
instance_name => 'inst1' );
END;
BEGIN
DBMS_MONITOR.SERV_MOD_ACT_TRACE_DISABLE(
service_name => 'ACCTG' ,
module_name => 'PAYROLL' ,
instance_name => 'inst1' );
END;
BEGIN
DBMS_SESSION.SESSION_TRACE_ENABLE(
waits => true
, binds => false);
END;
Assumptions
This tutorial assumes the following:
• You want to log in to the database with administrator privileges.
• User OE has one active session.
• You want to temporarily enable tracing for the OE session.
• You want to include wait information in the trace.
23-9
Chapter 23
Enabling End-to-End Application Tracing
3. Use the values from the preceding step to enable tracing for a specific session.
For example, execute the following program to enable tracing for the OE session,
where the true argument includes wait information in the trace and the false
argument excludes bind information from the trace:
BEGIN
DBMS_MONITOR.SESSION_TRACE_ENABLE(
session_id => 27
, serial_num => 60
, waits => true
, binds => false);
END;
BEGIN
DBMS_MONITOR.SESSION_TRACE_DISABLE(
session_id => 27
, serial_num => 60);
END;
23-10
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
Prerequisites
You must have administrative privileges to execute the DATABASE_TRACE_ENABLE
procedure.
Assumptions
This tutorial assumes the following:
• You want to enable tracing for all SQL the inst1 instance.
• You want wait information to be in the trace.
• You do not want bind information in the trace.
BEGIN
DBMS_MONITOR.DATABASE_TRACE_ENABLE(
waits => true
, binds => false
, instance_name => 'inst1' );
END;
To disable the session-level SQL tracing for an entire database, invoke the
DATABASE_TRACE_DISABLE procedure without specifying the instance_name
parameter:
EXECUTE DBMS_MONITOR.DATABASE_TRACE_DISABLE();
23-11
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
Parameter Description
DIAGNOSTIC_DEST Specifies the location of the Automatic Diagnostic
Repository (ADR) Home. The diagnostic files for each
database instance are located in this dedicated directory.
23-12
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
Parameter Description
MAX_DUMP_FILE_SIZE When the SQL Trace facility is enabled at the database
instance level, every call to the database writes a text line
in a file in the operating system's file format. The maximum
size of these files in operating system blocks is limited by
this initialization parameter. The default is UNLIMITED.
TIMED_STATISTICS Enables and disables the collection of timed statistics, such
as CPU and elapsed times, by the SQL Trace facility, and
the collection of various statistics in the V$ views.
If STATISTICS_LEVEL is set to TYPICAL or ALL, then the
default value of TIMED_STATISTICS is true. If
STATISTICS_LEVEL is set to BASIC, then the default value
of TIMED_STATISTICS is false.
Enabling timing causes extra timing calls for low-level
operations. This is a dynamic parameter. It is also a
session parameter.
3. If the operating system retains multiple versions of files, then ensure that the
version limit is high enough to accommodate the number of trace files you expect
the SQL Trace facility to generate.
4. If the generated trace files can be owned by an operating system user other than
yourself, then ensure that you have the necessary permissions to use TKPROF to
format them.
See Also:
23-13
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
Note:
Because running the SQL Trace facility increases system overhead,
enable it only when tuning SQL statements, and disable it when you are
finished.
23-14
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
DBMS_MONITOR.DATABASE_TRACE_ENABLE
Note:
The following SQL statements are truncated to 25 characters in the SQL
Trace file:
SET ROLE
GRANT
ALTER USER
ALTER ROLE
CREATE USER
CREATE ROLE
23-15
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
23-16
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
DATE_OF_INSERT DATE,
CURSOR_NUM NUMBER,
DEPTH NUMBER,
USER_ID NUMBER,
PARSE_CNT NUMBER,
PARSE_CPU NUMBER,
PARSE_ELAP NUMBER,
PARSE_DISK NUMBER,
PARSE_QUERY NUMBER,
PARSE_CURRENT NUMBER,
PARSE_MISS NUMBER,
EXE_COUNT NUMBER,
EXE_CPU NUMBER,
EXE_ELAP NUMBER,
EXE_DISK NUMBER,
EXE_QUERY NUMBER,
EXE_CURRENT NUMBER,
EXE_MISS NUMBER,
EXE_ROWS NUMBER,
FETCH_COUNT NUMBER,
FETCH_CPU NUMBER,
FETCH_ELAP NUMBER,
FETCH_DISK NUMBER,
FETCH_QUERY NUMBER,
FETCH_CURRENT NUMBER,
FETCH_ROWS NUMBER,
23-17
Chapter 23
Generating Output Files Using SQL Trace and TKPROF
CLOCK_TICKS NUMBER,
SQL_STATEMENT LONG);
Most output table columns correspond directly to the statistics that appear in the
formatted output file. For example, the PARSE_CNT column value corresponds to the
count statistic for the parse step in the output file.
The columns in the following table help you identify a row of statistics.
Column Description
SQL_STATEMENT This is the SQL statement for which the SQL Trace facility collected
the row of statistics. Because this column has data type LONG, you
cannot use it in expressions or WHERE clause conditions.
DATE_OF_INSERT This is the date and time when the row was inserted into the table.
This value is different from the time when the SQL Trace facility
collected the statistics.
DEPTH This indicates the level of recursion at which the SQL statement was
issued. For example, a value of 0 indicates that a user issued the
statement. A value of 1 indicates that Oracle Database generated the
statement as a recursive call to process a statement with a value of 0
(a statement issued by a user). A value of n indicates that Oracle
Database generated the statement as a recursive call to process a
statement with a value of n-1.
USER_ID This identifies the user issuing the statement. This value also appears
in the formatted output file.
CURSOR_NUM Oracle Database uses this column value to keep track of the cursor to
which each SQL statement was assigned.
The output table does not store the statement's execution plan. The following query
returns the statistics from the output table. These statistics correspond to the formatted
output shown in "Example 23-7".
23-18
Chapter 23
Guidelines for Interpreting TKPROF Output
SQL_STATEMENT
---------------------------------------------------------------------
SELECT * FROM EMP, DEPT WHERE EMP.DEPTNO = DEPT.DEPTNO
23-19
Chapter 23
Guidelines for Interpreting TKPROF Output
process a SQL statement, consider the statistics for that statement and those for
recursive calls caused by that statement.
Note:
Recursive SQL statistics are not included for SQL-level operations.
With the exception of locking problems and inefficient PL/SQL loops, neither the CPU
time nor the elapsed time is necessary to find problem statements. The key is the
number of block visits, both query (that is, subject to read consistency) and current
(that is, not subject to read consistency). Segment headers and blocks that are going
to be updated are acquired in current mode, but all query and subquery processing
requests the data in query mode. These are precisely the same measures as the
instance statistics CONSISTENT GETS and DB BLOCK GETS. You can find high disk
activity in the disk column.
The following listing shows TKPROF output for one SQL statement as it appears in the
output file:
SELECT *
FROM emp, dept
WHERE emp.deptno = dept.deptno;
If it is acceptable to have 7.01 CPU seconds and to retrieve 824 rows, then you need
not look any further at this trace output. In fact, a major use of TKPROF reports in a
tuning exercise is to eliminate processes from the detailed tuning phase.
The output indicates that 10 unnecessary parse call were made (because 11 parse
calls exist for this single statement) and that array fetch operations were performed.
More rows were fetched than there were fetches performed. A large gap between CPU
and elapsed timings indicates Physical I/Os.
23-20
Chapter 23
Guidelines for Interpreting TKPROF Output
See Also:
"Example 23-4"
SELECT name_id
FROM cq_names
WHERE name = 'FLOOR';
23-21
Chapter 23
Guidelines for Interpreting TKPROF Output
SELECT name_id
FROM cq_names
WHERE name = 'FLOOR';
Two statistics suggest that the query might have been executed with a full table scan:
the current mode block visits, and the number of rows originating from the Table
Access row source in the plan. The explanation is that the required index was built
after the trace file had been produced, but before TKPROF had been run. Generating a
new trace file gives the following data:
SELECT name_id
FROM cq_names
WHERE name = 'FLOOR';
23-22
Chapter 23
Application Tracing Utilities
In the correct version, the parse call took 10 milliseconds of CPU time and 20
milliseconds of elapsed time, but the query apparently took no time to execute and
perform the fetch. These anomalies arise because the clock tick of 10 milliseconds is
too long relative to the time taken to execute and fetch the data. In such cases, it is
important to get many executions of the statements, so that you have statistically valid
numbers.
UPDATE cq_names
SET ATTRIBUTES = lower(ATTRIBUTES)
WHERE ATTRIBUTES = :att
23-23
Chapter 23
Application Tracing Utilities
• TRCSESS
The TRCSESS utility consolidates trace output from selected trace files based on
user-specified criteria.
• TKPROF
The TKPROF program formats the contents of the trace file and places the output
into a readable output file.
23.6.1.1 TRCSESS
The TRCSESS utility consolidates trace output from selected trace files based on
user-specified criteria.
After TRCSESS merges the trace information into a single output file, TKPROF can
process the output file. This section contains the following topics:
• Purpose
TRCSESS is useful for consolidating the tracing of a particular session for
performance or debugging purposes.
• Guidelines
You must specify one of the session, clientid, service, action, or module
options.
• Syntax
• Options
TRCSESS supports a number of command-line options.
• Examples
This section demonstrates common TRCSESS use cases.
23.6.1.1.1 Purpose
TRCSESS is useful for consolidating the tracing of a particular session for
performance or debugging purposes.
Tracing a specific session is usually not a problem in the dedicated server model
because one process serves a session during its lifetime. You can see the trace
information for the session from the trace file belonging to the server process.
However, in a shared server configuration, a user session is serviced by different
processes over time. The trace for the user session is scattered across different trace
files belonging to different processes, which makes it difficult to get a complete picture
of the life cycle of a session.
23.6.1.1.2 Guidelines
You must specify one of the session, clientid, service, action, or module options.
If you specify multiple options, then TRCSESS consolidates all trace files that satisfy
the specified criteria into the output file.
23.6.1.1.3 Syntax
trcsess [output=output_file_name]
[session=session_id]
[clientid=client_id]
23-24
Chapter 23
Application Tracing Utilities
[service=service_name]
[action=action_name]
[module=module_name]
[trace_files]
23.6.1.1.4 Options
TRCSESS supports a number of command-line options.
Argument Description
output Specifies the file where the output is generated. If this option is not
specified, then the utility writes to standard output.
session Consolidates the trace information for the session specified. The
session identifier is a combination of session index and session serial
number, such as 21.2371. You can locate these values in the
V$SESSION view.
clientid Consolidates the trace information for the specified client ID.
service Consolidates the trace information for the specified service name.
action Consolidates the trace information for the specified action name.
module Consolidates the trace information for the specified module name.
trace_files Lists the trace file names, separated by spaces, in which TRCSESS
should look for trace information. You can use the wildcard character
(*) to specify the trace file names. If you do not specify trace files, then
TRCSESS uses all files in the current directory as input.
23.6.1.1.5 Examples
This section demonstrates common TRCSESS use cases.
Example 23-2 Tracing a Single Session
This sample output of TRCSESS shows the container of traces for a particular
session. In this example, the session index and serial number equals 21.2371. All files
in current directory are taken as input.
trcsess session=21.2371
[PROCESS ID = 12359]
*** 2014-04-02 09:48:28.376
PARSING IN CURSOR #1 len=17 dep=0 uid=27 oct=3 lid=27 tim=868373970961
hv=887450622 ad='22683fb4'
select * from cat
23-25
Chapter 23
Application Tracing Utilities
END OF STMT
PARSE #1:c=0,e=339,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=4,tim=868373970944
EXEC #1:c=0,e=221,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=4,tim=868373971411
FETCH #1:c=0,e=791,p=0,cr=7,cu=0,mis=0,r=1,dep=0,og=4,tim=868373972435
FETCH #1:c=0,e=1486,p=0,cr=20,cu=0,mis=0,r=6,dep=0,og=4,tim=868373986238
*** 2014-04-02 10:03:58.058
XCTEND rlbk=0, rd_only=1
STAT #1 id=1 cnt=7 pid=0 pos=1 obj=0 op='FILTER '
STAT #1 id=2 cnt=7 pid=1 pos=1 obj=18 op='TABLE ACCESS BY INDEX ROWID OBJ$
'
STAT #1 id=3 cnt=7 pid=2 pos=1 obj=37 op='INDEX RANGE SCAN I_OBJ2 '
STAT #1 id=4 cnt=0 pid=1 pos=2 obj=4 op='TABLE ACCESS CLUSTER TAB$J2 '
STAT #1 id=5 cnt=6 pid=4 pos=1 obj=3 op='INDEX UNIQUE SCAN I_OBJ# '
[PROCESS ID=12995]
*** 2014-04-02 10:04:32.738
Archiving is disabled
23.6.1.2 TKPROF
The TKPROF program formats the contents of the trace file and places the output into
a readable output file.
TKPROF can also do the following:
• Create a SQL script that stores the statistics in the database
• Determine the execution plans of SQL statements
Note:
If the cursor for a SQL statement is not closed, then TKPROF output
does not automatically include the actual execution plan of the SQL
statement. In this situation, use the EXPLAIN option with TKPROF to
generate an execution plan.
TKPROF reports each statement executed with the resources it has consumed, the
number of times it was called, and the number of rows which it processed.
This section contains the following topics:
• Purpose
TKPROF can locate statements that are consuming the greatest resources.
• Guidelines
The input and output files are the only required arguments.
• Syntax
• Options
TKPROF supports a number of command-line options.
• Output
This section explains the TKPROF output.
• Examples
This section demonstrates common TKPROF use cases.
23-26
Chapter 23
Application Tracing Utilities
23.6.1.2.1 Purpose
TKPROF can locate statements that are consuming the greatest resources.
With baselines available, you can assess whether the resources used are reasonable
given the work performed.
23.6.1.2.2 Guidelines
The input and output files are the only required arguments.
If you invoke TKPROF without arguments, then the tool displays online help.
23.6.1.2.3 Syntax
23.6.1.2.4 Options
TKPROF supports a number of command-line options.
Argument Description
input_file Specifies the input file, a trace file containing statistics produced by the
SQL Trace facility. This file can be either a trace file produced for a single
session, or a file produced by concatenating individual trace files from
multiple sessions.
output_file Specifies the file to which TKPROF writes its formatted output.
WAITS Specifies whether to record summary for any wait events found in the
trace file. Valid values are YES (default) and NO.
23-27
Chapter 23
Application Tracing Utilities
Argument Description
SORT Sorts traced SQL statements in descending order of specified sort option
before listing them in the output file. If multiple options are specified, then
the output is sorted in descending order by the sum of the values specified
in the sort options. If you omit this parameter, then TKPROF lists
statements into the output file in order of first use. Sort options are listed
as follows:
• PRSCNT - Number of times parsed
• PRSCPU - CPU time spent parsing
• PRSELA - Elapsed time spent parsing
• PRSDSK - Number of physical reads from disk during parse
• PRSQRY - Number of consistent mode block reads during parse
• PRSCU - Number of current mode block reads during parse
• PRSMIS - Number of library cache misses during parse
• EXECNT - Number of executions
• EXECPU - CPU time spent executing
• EXEELA - Elapsed time spent executing
• EXEDSK - Number of physical reads from disk during execute
• EXEQRY - Number of consistent mode block reads during execute
• EXECU - Number of current mode block reads during execute
• EXEROW - Number of rows processed during execute
• EXEMIS - Number of library cache misses during execute
• FCHCNT - Number of fetches
• FCHCPU - CPU time spent fetching
• FCHELA - Elapsed time spent fetching
• FCHDSK - Number of physical reads from disk during fetch
• FCHQRY - Number of consistent mode block reads during fetch
• FCHCU - Number of current mode block reads during fetch
• FCHROW - Number of rows fetched
• USERID - ID of user that parsed the cursor
PRINT Lists only the first integer sorted SQL statements from the output file. If
you omit this parameter, then TKPROF lists all traced SQL statements. This
parameter does not affect the optional SQL script. The SQL script always
generates insert data for all traced SQL statements.
AGGREGATE If you specify AGGREGATE = NO, then TKPROF does not aggregate multiple
users of the same SQL text.
INSERT Creates a SQL script that stores the trace file statistics in the database.
TKPROF creates this script with the name filename3. This script creates a
table and inserts a row of statistics for each traced SQL statement into the
table.
SYS Enables and disables the listing of SQL statements issued by the user
SYS, or recursive SQL statements, into the output file. The default value of
YES causes TKPROF to list these statements. The value of NO causes
TKPROF to omit them. This parameter does not affect the optional SQL
script. The SQL script always inserts statistics for all traced SQL
statements, including recursive SQL statements.
23-28
Chapter 23
Application Tracing Utilities
Argument Description
TABLE Specifies the schema and name of the table into which TKPROF
temporarily places execution plans before writing them to the output file. If
the specified table exists, then TKPROF deletes all rows in the table, uses it
for the EXPLAIN PLAN statement (which writes more rows into the table),
and then deletes those rows. If this table does not exist, then TKPROF
creates it, uses it, and then drops it.
The specified user must be able to issue INSERT, SELECT, and DELETE
statements against the table. If the table does not exist, then the user
must also be able to issue CREATE TABLE and DROP TABLE statements.
This option enables multiple users to run TKPROF concurrently with the
same database user account in the EXPLAIN value. These users can
specify different TABLE values and avoid destructively interfering with each
other when processing the temporary plan table.
TKPROF supports the following combinations:
• The EXPLAIN parameter without the TABLE parameter
TKPROF uses the table PROF$PLAN_TABLE in the schema of the user
specified by the EXPLAIN parameter
• The TABLE parameter without the EXPLAIN parameter
TKPROF ignores the TABLE parameter.
If no plan table exists, then TKPROF creates the table PROF$PLAN_TABLE
and then drops it at the end.
EXPLAIN Determines the execution plan for each SQL statement in the trace file
and writes these execution plans to the output file. TKPROF also displays
the number of rows processed by each step of the execution plan.
TKPROF determines execution plans by issuing the EXPLAIN PLAN
statement after connecting to Oracle Database with the user and
password specified in this parameter. The specified user must have
CREATE SESSION system privileges. TKPROF takes longer to process a
large trace file if the EXPLAIN option is used.
Note: Trace files generated immediately after instance startup contain
data that reflects the activity of the startup process. In particular, they
reflect a disproportionate amount of I/O activity as caches in the system
global area (SGA) are filled. For the purposes of tuning, ignore such trace
files.
RECORD Creates a SQL script with the specified filename with all of the
nonrecursive SQL in the trace file. You can use this script to replay the
user events from the trace file.
WIDTH An integer that controls the output line width of some TKPROF output, such
as the explain plan. This parameter is useful for post-processing of
TKPROF output.
23.6.1.2.5 Output
This section explains the TKPROF output.
This section contains the following topics:
23-29
Chapter 23
Application Tracing Utilities
If the SQL Trace input file contained statistics from multiple users, and if the statement
was issued by multiple users, then TKPROF lists the ID of the last user to parse the
statement. The user ID of all database users appears in the data dictionary in the
column ALL_USERS.USER_ID.
The other columns of the SQL Trace facility output are combined statistics for all
parses, executions, and fetches of a statement. The sum of query and current is the
total number of buffers accessed, also called Logical I/Os (LIOs). See Table 23-5.
23-30
Chapter 23
Application Tracing Utilities
Table 23-5 SQL Trace Statistics for Parses, Executes, and Fetches.
Statistics about the processed rows appear in the ROWS column. The column shows the
number of rows processed by the SQL statement. This total does not include rows
processed by subqueries of the SQL statement. For SELECT statements, the number of
rows returned appears for the fetch step. For UPDATE, DELETE, and INSERT statements,
the number of rows processed appears for the execute step.
Note:
The row source counts are displayed when a cursor is closed. In SQL*Plus,
there is only one user cursor, so each statement executed causes the
previous cursor to be closed; therefore, the row source counts are displayed.
PL/SQL has its own cursor handling and does not close child cursors when
the parent cursor is closed. Exiting or reconnecting causes the counts to be
displayed.
23-31
Chapter 23
Application Tracing Utilities
In the following sample TKPROF output, note the cr, r, w, and time values under the
Row Source Operation column:
23-32
Chapter 23
Application Tracing Utilities
0.00
SQL*Net message from client 1 0.00
0.00
In addition, wait events are summed for the entire trace file at the end of the file.
To ensure that wait events information is written to the trace file for the session, run
the following SQL statement:
ALTER SESSION SET EVENTS '10046 trace name context forever, level 8';
23.6.1.2.6 Examples
This section demonstrates common TKPROF use cases.
Example 23-4 Printing the Most Resource-Intensive Statements
If you are processing a large trace file using a combination of SORT parameters and the
PRINT parameter, then you can produce a TKPROF output file containing only the
highest resource-intensive statements. The following statement prints the 10
statements in the trace file that have generated the most physical I/O:
This example is likely to be longer than a single line on the screen, and you might
need to use continuation characters, depending on the operating system.
Note the other parameters in this example:
• The EXPLAIN value causes TKPROF to connect as the user hr and use the EXPLAIN
PLAN statement to generate the execution plan for each traced SQL statement.
You can use this to get access paths and row source counts.
Note:
If the cursor for a SQL statement is not closed, then TKPROF output does
not automatically include the actual execution plan of the SQL statement.
In this situation, you can use the EXPLAIN option with TKPROF to generate
an execution plan.
• The TABLE value causes TKPROF to use the table temp_plan_table_a in the
schema scott as a temporary plan table.
23-33
Chapter 23
Application Tracing Utilities
• The INSERT value causes TKPROF to generate a SQL script named STOREA.SQL that
stores statistics for all traced SQL statements in the database.
• The SYS parameter with the value of NO causes TKPROF to omit recursive SQL
statements from the output file. In this way, you can ignore internal Oracle
Database statements such as temporary table operations.
• The SORT value causes TKPROF to sort the SQL statements in order of the sum of
the CPU time spent executing and the CPU time spent fetching rows before writing
them to the output file. For greatest efficiency, always use SORT parameters.
Example 23-6 TKPROF Header
This example shows a sample header for the TKPROF report.
Copyright (c) 1982, 2012, Oracle and/or its affiliates. All rights
reserved.
***************************************************************************
*****
count = number of times OCI procedure was executed
cpu = cpu time in seconds executing
elapsed = elapsed time in seconds executing
disk = number of physical reads of buffers from disk
query = number of buffers gotten for consistent read
current = number of buffers gotten in current mode (usually for update)
rows = number of rows processed by the fetch or execute call
***************************************************************************
*****
23-34
Chapter 23
Application Tracing Utilities
select condition
from
cdef$ where rowid=:1
***************************************************************************
*****
23-35
Chapter 23
Application Tracing Utilities
0 0
Execute 1 0.00 0.00 0 0
0 0
Fetch 2 0.00 0.00 0 15
0 1
------- ------ -------- ---------- ---------- ---------- ----------
----------
total 4 0.02 0.01 0 15
0 1
***************************************************************************
*****
delete
from stats$sqltext st
where (hash_value, text_subset) not in
(select --+ hash_aj
hash_value, text_subset
from stats$sql_summary ss
where ( ( snap_id < :lo_snap
or snap_id > :hi_snap
)
and dbid = :dbid
and instance_number = :inst_num
)
or ( dbid != :dbid
or instance_number != :inst_num)
)
23-36
Chapter 23
Application Tracing Utilities
23-37
Chapter 23
Application Tracing Utilities
23-38
Chapter 23
Views for Application Tracing
*****
Trace file: main_ora_27621.trc
Trace file compatibility: 9.00.01
Sort options: default
1 session in tracefile.
5 user SQL statements in trace file.
1 internal SQL statements in trace file.
6 SQL statements in trace file.
6 unique SQL statements in trace file.
76 lines in trace file.
128 elapsed seconds in trace file.
View Description
DBA_ENABLED_AGGREGATIONS Accumulated global statistics for the currently enabled statistics
V$CLIENT_STATS Accumulated statistics for a specified client identifier
V$SERVICE_STATS Accumulated statistics for a specified service
V$SERV_MOD_ACT_STATS Accumulated statistics for a combination of specified service,
module, and action
V$SERVICEMETRIC Accumulated statistics for elapsed time of database calls and
for CPU use
V$DIAG_TRACE_FILE Information about all trace files in ADR for the current container
V$DIAG_APP_TRACE_FILE Information about all trace files that contain application trace
data (SQL_TRACE or OPTIMIZER_TRACE event data) in ADR for
the current container
V$DIAG_TRACE_FILE_CONTENTS Trace data in the trace files in ADR
V$DIAG_SQL_TRACE_RECORDS SQL_TRACE data in the trace files in ADR
V$DIAG_OPT_TRACE_RECORDS Optimizer trace event data in the trace files in ADR
V$DIAG_SESS_SQL_TRACE_RECORDS SQL_TRACE data in the trace files in ADR for the current user
session
23-39
Chapter 23
Views for Application Tracing
View Description
V$DIAG_SESS_OPT_TRACE_RECORDS Optimizer trace event data in the trace files in ADR for the
current user session
V$DIAG_ALERT_EXT Contents of the XML-based alert log in ADR for the current
container
See Also:
Oracle Database Reference for information about V$ and data dictionary
views
23-40
Part VIII
Automatic SQL Tuning
SQL Tuning Advisor and SQL Access Advisor are built-in tools that provide SQL
tuning recommendations.
This part contains the following chapters:
• Managing SQL Tuning Sets
You can use SQL tuning sets to group statements and related metadata into a
single object, which you can use as input to SQL tuning tools.
• Analyzing SQL with SQL Tuning Advisor
Use SQL Tuning Advisor to obtain recommendations for improving performance of
high-load SQL statements, and prevent regressions by only executing optimal
plans.
• Optimizing Access Paths with SQL Access Advisor
SQL Access Advisor is diagnostic software that identifies and helps resolve SQL
performance problems by recommending indexes, materialized views,
materialized view logs, or partitions to create, drop, or retain.
24
Managing SQL Tuning Sets
You can use SQL tuning sets to group statements and related metadata into a single
object, which you can use as input to SQL tuning tools.
This chapter contains the following topics:
• About SQL Tuning Sets
A SQL tuning set (STS) is a database object that you can use as input to tuning
tools.
• Creating a SQL Tuning Set
Use the DBMS_SQLTUNE.CREATE_SQLSET procedure to create an empty STS in the
database.
• Loading a SQL Tuning Set
To load an STS with SQL statements, execute the DBMS_SQLTUNE.LOAD_SQLSET
procedure.
• Displaying the Contents of a SQL Tuning Set
After an STS has been created and populated, execute the
DBMS_SQLTUNE.SELECT_SQLSET function to read the contents of the STS, optionally
using filtering criteria.
• Modifying a SQL Tuning Set
Use the DBMS_SQLTUNE.DELETE_SQLSET procedure to delete SQL statements from
an STS.
• Transporting a SQL Tuning Set
You can transport an STS to any database created in Oracle Database 10g
Release 2 (10.2) or later. This technique is useful when using SQL Performance
Analyzer to tune regressions on a test database.
• Dropping a SQL Tuning Set
To drop an STS from the database, execute the DBMS_SQLTUNE.DROP_SQLSET
procedure.
24-1
Chapter 24
About SQL Tuning Sets
• Associated execution plans and row source statistics for each SQL statement
(optional)
Note:
Data visibility and privilege requirements may differ when using an STS with
pluggable databases. See Oracle Database Administrator’s Guide for a table
that summarizes how manageability features work in a container database
(CDB).
24-2
Chapter 24
About SQL Tuning Sets
SQL Tuning
Advisor
Custom
SQL
SQL Access
Advisor
AWR
Shared SQL
Area Filter
STS
Transport
See Also:
Oracle Database Performance Tuning Guide to learn about AWR
24-3
Chapter 24
About SQL Tuning Sets
See Also:
Oracle Database 2 Day + Performance Tuning Guide
24-4
Chapter 24
About SQL Tuning Sets
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
DBMS_SQLTUNE
Create STS
CREATE_SQLSET
Populate STS
LOAD_SQLSET with SQL
CREATE_STGTAB_SQLSET
Optionally, transport
PACK_STGTAB_SQLSET STS to different
database
Transport STS
UNPACK_STGTAB_SQLSET
Drop STS
DROP_SQLSET
24-5
Chapter 24
Creating a SQL Tuning Set
Parameter Description
sqlset_name Name of the STS
description Optional description of the STS
Assumptions
This tutorial assumes that you want to create an STS named SQLT_WKLD_STS.
To create an STS:
1. Connect SQL*Plus to the database with the appropriate privileges, and then run
the DBMS_SQLTUNE.CREATE_SQLSET procedure.
For example, execute the following PL/SQL program:
BEGIN
DBMS_SQLTUNE.CREATE_SQLSET (
sqlset_name => 'SQLT_WKLD_STS'
, description => 'STS to store SQL from the private SQL area'
);
END;
24-6
Chapter 24
Loading a SQL Tuning Set
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information
Parameter Description
populate_cursor Specifies the cursor reference from which to populate the STS.
load_option Specifies how the statements are loaded into the STS. The
possible values are INSERT (default), UPDATE, and MERGE.
24-7
Chapter 24
Loading a SQL Tuning Set
high-load SQL statements, or the LOAD_SQLSET procedure, which accesses the data
source only once.
Prerequisites
This tutorial has the following prerequisites:
• Filters provided to the SELECT_CURSOR_CACHE function are evaluated as part of SQL
statements run by the current user. As such, they are executed with that user's
security privileges and can contain any constructs and subqueries that user can
access, but no more.
• The current user must have privileges on the shared SQL area views.
Assumptions
This tutorial assumes that you want to load the SQL tuning set named SQLT_WKLD_STS
with statements from the shared SQL area.
To load an STS:
1. Connect SQL*Plus to the database as a user with the appropriate privileges.
2. Run the DBMS_SQLTUNE.LOAD_SQLSET procedure.
For example, execute the following PL/SQL program to populate a SQL tuning set
with all cursor cache statements that belong to the sh schema:
DECLARE
c_sqlarea_cursor DBMS_SQLTUNE.SQLSET_CURSOR;
BEGIN
OPEN c_sqlarea_cursor FOR
SELECT VALUE(p)
FROM TABLE(
DBMS_SQLTUNE.SELECT_CURSOR_CACHE(
' module = ''SQLT_WKLD'' AND parsing_schema_name = ''SH'' ')
) p;
-- load the tuning set
DBMS_SQLTUNE.LOAD_SQLSET (
sqlset_name => 'SQLT_WKLD_STS'
, populate_cursor => c_sqlarea_cursor
);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information.
24-8
Chapter 24
Displaying the Contents of a SQL Tuning Set
Parameter Description
basic_filter The SQL predicate to filter the SQL from the STS defined on attributes
of the SQLSET_ROW
object_filter Specifies the objects that exist in the object list of selected SQL from
the shared SQL area
Table 24-4 describes some attributes of the SQLSET_ROW object. These attributes
appears as columns when you query TABLE(DBMS_SQLTUNE.SELECT_SQLSET()).
Parameter Description
parsing_schema_na Schema in which the SQL is parsed
me
elapsed_time Sum of the total number of seconds elapsed for this SQL statement
buffer_gets Total number of buffer gets (number of times the database accessed a
block) for this SQL statement
Assumptions
This tutorial assumes that you want to display the contents of an STS named
SQLT_WKLD_STS.
24-9
Chapter 24
Displaying the Contents of a SQL Tuning Set
24-10
Chapter 24
Modifying a SQL Tuning Set
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information
Assumptions
This tutorial assumes that you want to modify SQLT_WKLD_STS as follows:
• You want to delete all SQL statements with fetch counts over 100.
• You want to change the priority of the SQL statement with ID fudq5z56g642p to 1.
You can use priority as a ranking criteria when running SQL Tuning Advisor.
24-11
Chapter 24
Modifying a SQL Tuning Set
Use the basic_filter predicate to filter the SQL from the STS defined on
attributes of the SQLSET_ROW. The following example deletes all statements in the
STS with fetch counts over 100:
BEGIN
DBMS_SQLTUNE.DELETE_SQLSET (
sqlset_name => 'SQLT_WKLD_STS'
, basic_filter => 'fetches > 100'
);
END;
/
BEGIN
DBMS_SQLTUNE.UPDATE_SQLSET (
sqlset_name => 'SQLT_WKLD_STS'
, sql_id => '2cqsw036j5u7r'
, attribute_name => 'PRIORITY'
, attribute_value => 1
);
END;
/
4. Optionally, query the STS to confirm that the intended modifications were made.
For example, execute the following query:
See Also:
Oracle Database PL/SQL Packages and Types Reference for more
information
24-12
Chapter 24
Transporting a SQL Tuning Set
24-13
Chapter 24
Transporting a SQL Tuning Set
Production Test
Database Database
24.6.1.2 Basic Steps for Transporting SQL Tuning Sets When the CON_DBID
Values Differ
When transporting an STS, you must remap the con_dbid of each SQL statement in
the STS when the con_dbid of the source database and the destination database are
different.
Situations that cause the con_dbid value to differ include the following:
24-14
Chapter 24
Transporting a SQL Tuning Set
BEGIN
DBMS_SQLTUNE.REMAP_STGTAB_SQLSET (
staging_table_name => 'src_stg_tbl'
, staging_schema_owner => 'dba1'
, old_sqlset_name => 'STS_for_transport'
, old_con_dbid => :con_dbid_src
, new_con_dbid => 12345);
END;
You can now export the contents of the staging table, and then continue using the
normal transport procedure.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
REMAP_STGTAB_SQLSET
Procedure Description
CREATE_STGTAB_SQLSET Create a staging table to hold the exported SQL tuning
sets
24-15
Chapter 24
Transporting a SQL Tuning Set
Procedure Description
PACK_STGTAB_SQLSET Populate a staging table with SQL tuning sets
UNPACK_STGTAB_SQLSET Copy the SQL tuning sets from the staging table into a
database
Assumptions
This tutorial assumes the following:
• An STS with regressed SQL resides in a production database created in the
current release.
• You run SQL Performance Analyzer trials on a remote test database created in
Oracle Database 11g Release 2 (11.2).
• You want to copy the STS from the production database to the test database and
tune the regressions from the SQL Performance Analyzer trials.
• You want to use Oracle Database Pump to transfer the SQL tuning sets between
database hosts.
To transport an STS:
1. Connect SQL*Plus to the production database with administrator privileges.
2. Use the CREATE_STGTAB_SQLSET procedure to create a staging table to hold the
exported SQL tuning sets.
The following example creates my_11g_staging_table in the dba1 schema and
specifies the format of the staging table as 11.2:
BEGIN
DBMS_SQLTUNE.CREATE_STGTAB_SQLSET (
table_name => 'my_10g_staging_table'
, schema_name => 'dba1'
, db_version => DBMS_SQLTUNE.STS_STGTAB_11_2_VERSION
);
END;
/
3. Use the PACK_STGTAB_SQLSET procedure to populate the staging table with SQL
tuning sets.
The following example populates dba1.my_11g_staging_table with the STS
my_sts owned by hr:
BEGIN
DBMS_SQLTUNE.PACK_STGTAB_SQLSET (
sqlset_name => 'sqlt_wkld_sts'
, sqlset_owner => 'sh'
, staging_table_name => 'my_11g_staging_table'
, staging_schema_owner => 'dba1'
, db_version => DBMS_SQLTUNE.STS_STGTAB_11_2_VERSION
24-16
Chapter 24
Dropping a SQL Tuning Set
);
END;
/
4. If necessary, remap the container ID values for the statements in the STS as
described in "Basic Steps for Transporting SQL Tuning Sets When the CON_DBID
Values Differ".
5. Use Oracle Data Pump to export the contents of the staging table.
For example, run the expdp command at the operating system prompt:
BEGIN
DBMS_SQLTUNE.UNPACK_STGTAB_SQLSET (
sqlset_name => '%'
, replace => true
, staging_table_name => 'my_11g_staging_table');
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about DBMS_SQLTUNE.UNPACK_STGTAB_SQLSET
Prerequisites
Ensure that no tuning task is currently using the STS to be dropped. If an existing
tuning task is using this STS, then drop the task before dropping the STS. Otherwise,
the database issues an ORA-13757 error.
24-17
Chapter 24
Dropping a SQL Tuning Set
Assumptions
This tutorial assumes that you want to drop an STS named SQLT_WKLD_STS.
To drop an STS:
1. Start SQL*Plus, and then log in to the database with the appropriate privileges.
2. Run the DBMS_SQLTUNE.DROP_SQLSET procedure.
For example, execute the following PL/SQL program:
BEGIN
DBMS_SQLTUNE.DROP_SQLSET( sqlset_name => 'SQLT_WKLD_STS' );
END;
/
SELECT COUNT(*)
FROM USER_SQLSET
WHERE NAME = 'SQLT_WKLD_STS';
COUNT(*)
----------
0
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
STS procedures in DBMS_SQLTUNE
24-18
25
Analyzing SQL with SQL Tuning Advisor
Use SQL Tuning Advisor to obtain recommendations for improving performance of
high-load SQL statements, and prevent regressions by only executing optimal plans.
This chapter contains the following topics:
• About SQL Tuning Advisor
SQL Tuning Advisor is SQL diagnostic software in the Oracle Database Tuning
Pack.
• Managing the Automatic SQL Tuning Task
When your goal is to identify SQL performance problems proactively, configuring
SQL Tuning Advisor as an automated task is a simple solution. The task
processes selected high-load SQL statements from AWR that qualify as tuning
candidates.
• Running SQL Tuning Advisor On Demand
You can run SQL Tuning Advisor on demand.
25-1
Chapter 25
About SQL Tuning Advisor
Note:
Data visibility and privilege requirements may differ when using SQL Tuning
Advisor with pluggable databases. The advisor can tune a query in the
current pluggable database (PDB), and in other PDBs in which this query
has been executed. In this way, a container database (CDB) administrator
can tune the same query in many PDBs at the same time, whereas a PDB
administrator can only tune a single PDB.
See Also:
25-2
Chapter 25
About SQL Tuning Advisor
Optimizer
ADDM Automatic
Tuning
Optimizer
AWR
SQL Tuning
Advisor
Shared Pool
Inplementation
Shared SQL Area of SQL Profiles
(Automatic Only)
SELECT * FROM
employees
SQL AUTOTASK
Tuning
Set
25-3
Chapter 25
About SQL Tuning Advisor
See Also:
"SQL Parsing"
25-4
Chapter 25
About SQL Tuning Advisor
See Also:
For example, assume that before tuning the execution time was 100 seconds, and
after implementing the recommendation the new execution time is expected to be 33
seconds. This benefit calculation for this performance improvement is as follows:
You choose whether to accept the recommendations to optimize the SQL statements.
Depending on how it is configured, Automatic SQL Tuning Advisor can implement the
SQL profile recommendations to tune the statement without user intervention. When
invoked on demand, SQL Tuning Advisor can recommend that the user implement a
SQL profile, but can never implement it automatically.
25-5
Chapter 25
About SQL Tuning Advisor
Optimizer
Automatic Tuning
Optimizer
Normal Mode
Tuning Mode
SQL Tuning
Advisor
• Statistical Analysis
The optimizer relies on object statistics to generate execution plans.
• SQL Profiling
SQL profiling is the verification by the Automatic Tuning Optimizer of its own
estimates.
• Access Path Analysis
An access path is the means by which the database retrieves data.
• SQL Structural Analysis
During structural analysis, Automatic Tuning Optimizer tries to identify syntactic,
semantic, or design problems that can lead to suboptimal performance. The goal
is to identify poorly written SQL statements and to advise you how to restructure
them.
• Alternative Plan Analysis
While tuning a SQL statement, SQL Tuning Advisor searches real-time and
historical performance data for alternative execution plans for the statement.
See Also:
"Query Optimizer Concepts "
25-6
Chapter 25
About SQL Tuning Advisor
SELECT . . .
SQL Tuning
Advisor
Optimizer
Automatic Tuning
Optimizer
Customers Customers
Table Table
Stale
Statistics Absent
Recommended collecting
object-level statistics Statistics
25-7
Chapter 25
About SQL Tuning Advisor
SQL Tuning
Advisor SQL
Profile
Optimizer
(Tuning Mode)
Submit Create
Use
Database
Users
Optimizer
No application GB Well-Tuned
(Normal Mode)
code change Output Plan
HJ
HJ
See Also:
"About SQL Profiles"
25-8
Chapter 25
About SQL Tuning Advisor
After performing its analysis, SQL Tuning Advisor either recommends or does not
recommend implementing a SQL profile.
The following graphic shows the SQL profiling process.
SELECT . . .
SQL Tuning
Advisor
Optimizer
Automatic Tuning
Optimizer
SQL Profiling
* Reviews past execution history to
adjust settings
* Performs sampling or partial
execution
Recommendation
No Recommendation to Implement
SQL Profile
During SQL profiling, the optimizer verifies cost, selectivity, and cardinality for a
statement. The optimizer uses either of the following methods:
• Samples the data and applies appropriate predicates to the sample
The optimizer compares the new estimate to the regular estimate and, if the
difference is great enough, applies a correction factor.
• Executes a fragment of the SQL statement
This method is more efficient than the sampling method when the predicates
provide efficient access paths.
The optimizer uses the past statement execution history to determine correct settings.
For example, if the history indicates that a SQL statement is usually executed only
partially, then the optimizer uses FIRST_ROWS instead of ALL_ROWS optimization.
See Also:
"Choosing an Optimizer Goal"
25-9
Chapter 25
About SQL Tuning Advisor
Note:
The Automatic SQL Tuning task cannot automatically create SQL plan
baselines or add plans to them.
User must
choose whether
SQL Tuning On Demand to implement
Advisor
Recommends
Implementing
SQL profile No Autoimplementation
Autotask
Implements
SQL profile
Autoimplementation
At any time during or after automatic SQL tuning, you can view a report. This report
describes in detail the SQL statements that were analyzed, the recommendations
generated, and any SQL profiles that were automatically implemented.
25-10
Chapter 25
About SQL Tuning Advisor
See Also:
• "Configuring the Automatic SQL Tuning Task Using the Command Line"
• "Plan Evolution"
• "About SQL Profiles"
• Oracle Database PL/SQL Packages and Types Reference to learn more
about ACCEPT_SQL_PROFILE
SELECT . . .
SQL Tuning
Advisor
Optimizer
Automatic Tuning
Optimizer
Recommends
Workload
SQL Access
Advisor
Index
Creation
Comprehensive
Analysis
Automatic Tuning Optimizer explores whether a new index can significantly enhance
query performance and recommends either of the following:
• Creating an index
25-11
Chapter 25
About SQL Tuning Advisor
SELECT . . . UNION
SQL Tuning
Advisor
Optimizer
Automatic Tuning
Optimizer
Restructured
25-12
Chapter 25
About SQL Tuning Advisor
25-13
Chapter 25
About SQL Tuning Advisor
Optimizer
Automatic Tuning
Optimizer
Real-Time
Performance Data
GB Origin:
Cursor
Cache SQL Tuning
HJ Advisor
HJ
Searches Produces
Alternative Plan Finding
AWR Performance Summary
Recommendations
GB Origin:
STS
HJ
HJ
SQL Tuning Advisor validates the alternative execution plans and notes any plans that
are not reproducible. When reproducible alternative plans are found, you can create a
SQL plan baseline to instruct the optimizer to choose these plans in the future.
Example 25-1 Alternative Plan Finding
The following example shows an alternative plan finding for a SELECT statement:
The following table lists these plans ranked by their average elapsed
time.
See section "ALTERNATIVE PLANS SECTION" for detailed information on each
plan.
Information
25-14
Chapter 25
About SQL Tuning Advisor
-----------
- The Original Plan appears to have the best performance, based on the
elapsed time per execution. However, if you know that one alternative
plan is better than the Original Plan, you can create a SQL plan
baseline
for it. This will instruct the Oracle optimizer to pick it over any
other
choices in the future.
execute dbms_sqltune.create_sql_plan_baseline(task_name =>
'TASK_XXXXX',
object_id => 2, task_owner => 'SYS', plan_hash => xxxxxxxx);
The preceding example shows that SQL Tuning Advisor found two plans, one in the
shared SQL area and one in a SQL tuning set. The plan in the shared SQL area is the
same as the original plan.
SQL Tuning Advisor only recommends an alternative plan if the elapsed time of the
original plan is worse than alternative plans. In this case, SQL Tuning Advisor
recommends that users create a SQL plan baseline on the plan with the best
performance. In Example 25-1, the alternative plan did not perform as well as the
original plan, so SQL Tuning Advisor did not recommend using the alternative plan.
Example 25-2 Alternative Plans Section
In this example, the alternative plans section of the SQL Tuning Advisor output
includes both the original and alternative plans and summarizes their performance.
The most important statistic is elapsed time. The original plan used an index, whereas
the alternative plan used a full table scan, increasing elapsed time by .002 seconds.
Plan 1
------
Notes:
1. Statistics shown are averaged over multiple executions.
2. The plan matches the original plan.
--------------------------------------------
| Id | Operation | Name |
--------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | SORT AGGREGATE | |
| 2 | MERGE JOIN | |
| 3 | INDEX FULL SCAN | TEST1_INDEX |
| 4 | SORT JOIN | |
| 5 | TABLE ACCESS FULL| TEST |
--------------------------------------------
25-15
Chapter 25
About SQL Tuning Advisor
Plan 2
------
Notes:
1. Statistics shown are averaged over multiple executions.
-------------------------------------
| Id | Operation | Name |
-------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | SORT AGGREGATE | |
| 2 | HASH JOIN | |
| 3 | TABLE ACCESS FULL| TEST |
| 4 | TABLE ACCESS FULL| TEST1 |
-------------------------------------
See Also:
"Differences Between SQL Plan Baselines and SQL Profiles"
25-16
Chapter 25
About SQL Tuning Advisor
See Also:
25-17
Chapter 25
About SQL Tuning Advisor
The following figure illustrates the general setup for tuning a standby database
workload on a primary database. This technique requires a standby-to-primary
database link.
The following table illustrates a typical remote tuning session. You issue all statements
on the standby database. DBMS_SQLTUNE uses the database link both to fetch data from
the primary database, and store data in the primary database.
Table 25-1 Tuning a Standby Database Workload Using a Database Link to the
Primary Database
25-18
Chapter 25
Managing the Automatic SQL Tuning Task
Table 25-1 (Cont.) Tuning a Standby Database Workload Using a Database Link
to the Primary Database
See Also:
See Also:
Oracle Database Administrator’s Guide to learn more about automated
maintenance tasks
25-19
Chapter 25
Managing the Automatic SQL Tuning Task
See Also:
Oracle Database Administrator’s Guide to learn about SQL Tuning Advisor in
a multitenant environment
25-20
Chapter 25
Managing the Automatic SQL Tuning Task
By default, the Automatic SQL Tuning task runs for at most one hour in a nightly
maintenance window. You can customize attributes of the maintenance windows,
including start and end time, frequency, and days of the week.
See Also:
Package Description
DBMS_AUTO_SQLTUNE Enables you run SQL Tuning Advisor, manage SQL
profiles, manage SQL tuning sets, and perform real-time
SQL performance monitoring. To use this API, you must
have the ADVISOR privilege.
DBMS_AUTO_TASK_ADMIN Provides an interface to AUTOTASK. You can use this
interface to enable and disable the Automatic SQL
Tuning task.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
DBMS_SQLTUNE ad DBMS_AUTO_TASK_ADMIN
25-21
Chapter 25
Managing the Automatic SQL Tuning Task
DBMS_SQLTUNE. DBMS_SQLTUNE.
SET_TUNING_TASK_PARAMETER REPORT_AUTO_TUNING_TASK
25-22
Chapter 25
Managing the Automatic SQL Tuning Task
25.2.2.1 Enabling and Disabling the Automatic SQL Tuning Task Using Cloud
Control
You can enable and disable all automatic maintenance tasks, including the Automatic
SQL Tuning task, using Cloud Control.
To enable or disable the Automatic SQL Tuning task using Cloud Control:
1. Log in to Cloud Control with the appropriate credentials.
2. Under the Targets menu, select Databases.
3. In the list of database targets, select the target for the Oracle Database instance
that you want to administer.
4. If prompted for database credentials, then enter the minimum credentials
necessary for the tasks you intend to perform.
5. From the Administration menu, select Oracle Scheduler, then Automated
Maintenance Tasks.
The Automated Maintenance Tasks page appears.
This page shows the predefined tasks. You access each task by clicking the
corresponding link to get more information about the task.
6. Click Automatic SQL Tuning.
The Automatic SQL Tuning Result Summary page appears.
The Task Status section shows whether the Automatic SQL Tuning Task is
enabled or disabled. In the following graphic, the task is disabled:
25-23
Chapter 25
Managing the Automatic SQL Tuning Task
25.2.2.2 Enabling and Disabling the Automatic SQL Tuning Task from the
Command Line
If you do not use Cloud Control to enable and disable the Automatic SQL Tuning task,
then you must use the command line.
You have the following options:
• Run the ENABLE or DISABLE procedure in the DBMS_AUTO_TASK_ADMIN PL/SQL
package.
This package is the recommended command-line technique. For both the ENABLE
and DISABLE procedures, you can specify a particular maintenance window with
the window_name parameter.
• Set the STATISTICS_LEVEL initialization parameter to BASIC to disable collection of
all advisories and statistics, including Automatic SQL Tuning Advisor.
Because monitoring and many automatic features are disabled, Oracle strongly
recommends that you do not set STATISTICS_LEVEL to BASIC.
BEGIN
DBMS_AUTO_TASK_ADMIN.ENABLE (
client_name => 'sql tuning advisor'
, operation => NULL
, window_name => NULL
);
END;
/
BEGIN
DBMS_AUTO_TASK_ADMIN.DISABLE (
client_name => 'sql tuning advisor'
, operation => NULL
, window_name => NULL
25-24
Chapter 25
Managing the Automatic SQL Tuning Task
);
END;
/
CLIENT_NAME STATUS
-------------------- --------
sql tuning advisor ENABLED
System altered.
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information
25-25
Chapter 25
Managing the Automatic SQL Tuning Task
25.2.3.1 Configuring the Automatic SQL Tuning Task Using Cloud Control
You can enable and disable all automatic maintenance tasks, including the Automatic
SQL Tuning task, using Cloud Control. You must perform the operation as SYS or have
the EXECUTE privilege on the PL/SQL package DBMS_AUTO_SQLTUNE.
25-26
Chapter 25
Managing the Automatic SQL Tuning Task
25.2.3.2 Configuring the Automatic SQL Tuning Task Using the Command Line
The DBMS_AUTO_SQLTUNE package enables you to configure automatic SQL tuning by
specifying the task parameters using the SET_AUTO_TUNING_TASK_PARAMETER
procedure.
Because the task is owned by SYS, only SYS can set task parameters.
Note:
When automatic implementation is enabled, the advisor only implements
recommendations to create SQL profiles. Recommendations such as
creating new indexes, gathering optimizer statistics, and creating SQL plan
baselines are not automatically implemented.
Assumptions
This tutorial assumes the following:
• You want the database to implement SQL profiles automatically, but to implement
no more than 50 SQL profiles per execution, and no more than 50 profiles total on
the database.
• You want the task to time out after 1200 seconds per execution.
PARAMETER_NAME VALUE
------------------------- ----------
25-27
Chapter 25
Managing the Automatic SQL Tuning Task
EXECUTION_DAYS_TO_EXPIRE 30
LOCAL_TIME_LIMIT 1000
ACCEPT_SQL_PROFILES FALSE
MAX_SQL_PROFILES_PER_EXEC 20
MAX_AUTO_SQL_PROFILES 10000
BEGIN
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER (
task_name => 'SYS_AUTO_SQL_TUNING_TASK'
, parameter => parameter_name
, value => value
);
END;
/
BEGIN
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER('SYS_AUTO_SQL_TUNING_TASK',
'LOCAL_TIME_LIMIT', 1200);
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER('SYS_AUTO_SQL_TUNING_TASK',
'ACCEPT_SQL_PROFILES', 'true');
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER('SYS_AUTO_SQL_TUNING_TASK',
'MAX_SQL_PROFILES_PER_EXEC', 50);
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER('SYS_AUTO_SQL_TUNING_TASK',
'MAX_AUTO_SQL_PROFILES', 10002);
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information for DBMS_AUTO_SQLTUNE
25-28
Chapter 25
Managing the Automatic SQL Tuning Task
statements tuned during the maintenance, and the number of SQL profiles
created.
• Summary
This section lists the SQL statements (by their SQL identifiers) that were tuned
during the maintenance window and the estimated benefit of each SQL profile, or
the execution statistics after performing a test execution of the SQL statement with
the SQL profile.
• Tuning findings
This section contains the following information about each SQL statement
analyzed by SQL Tuning Advisor:
– All findings associated with each SQL statement
– Whether the profile was implemented on the database, and why
– Whether the SQL profile is currently enabled on the database
– Detailed execution statistics captured when testing the SQL profile
• Explain plans
This section shows the old and new explain plans used by each SQL statement
analyzed by SQL Tuning Advisor.
• Errors
This section lists all errors encountered by the automatic SQL tuning task.
This section contains the following topics:
• Viewing Automatic SQL Tuning Reports Using the Command Line
To generate a SQL tuning report as a CLOB, execute the
DBMS_SQLTUNE.REPORT_AUTO_TUNING_TASK function.
25.2.4.1 Viewing Automatic SQL Tuning Reports Using the Command Line
To generate a SQL tuning report as a CLOB, execute the
DBMS_SQLTUNE.REPORT_AUTO_TUNING_TASK function.
You can store the CLOB in a variable and then print the variable to view the report.
Assumptions
This section assumes that you want to show all SQL statements that were analyzed in
the most recent execution, including recommendations that were not implemented.
25-29
Chapter 25
Managing the Automatic SQL Tuning Task
PRINT :my_rept
2. Read the general information section for an overview of the tuning execution.
The following sample shows the Automatic SQL Tuning task analyzed 17 SQL
statements in just over 7 minutes:
MY_REPT
------------------------------------------------------------------------
---
GENERAL INFORMATION SECTION
------------------------------------------------------------------------
---
Tuning Task Name : SYS_AUTO_SQL_TUNING_TASK
Tuning Task Owner : SYS
Workload Type : Automatic High-Load SQL
Workload
Execution Count : 6
Current Execution : EXEC_170
Execution Type : TUNE SQL
Scope : COMPREHENSIVE
Global Time Limit(seconds) : 3600
Per-SQL Time Limit(seconds) : 1200
Completion Status : COMPLETED
Started at : 04/16/2012 10:00:00
Completed at : 04/16/2012 10:07:11
Number of Candidate SQLs : 17
Cumulative Elapsed Time of SQL (s) : 8
------------------------------------------------------------------------
---
SQLs with Findings Ordered by Maximum (Profile/Index) Benefit,
Object ID
------------------------------------------------------------------------
---
ob ID SQL ID statistics profile(benefit) index(benefit)
restructure
25-30
Chapter 25
Managing the Automatic SQL Tuning Task
------------------------------------------------------------------------
---
DETAILS SECTION
------------------------------------------------------------------------
---
Statements with Results Ordered by Max (Profile/Index) Benefit, Object
ID
------------------------------------------------------------------------
---
Object ID : 82
Schema Name: DBA1
SQL ID : dqjcc345dd4ak
SQL Text : SELECT status FROM dba_autotask_client WHERE client_name=:1
------------------------------------------------------------------------
---
FINDINGS SECTION (1 finding)
------------------------------------------------------------------------
---
Validation results
------------------
The SQL profile was tested by executing both its plan and the
original
plan and measuring their respective execution statistics. A plan
may have been only partially executed if the other could be run
to completion in less time.
25-31
Chapter 25
Running SQL Tuning Advisor On Demand
Notes
-----
1. The original plan was first executed to warm the buffer cache.
2. Statistics for original plan were averaged over next 9 executions.
3. The SQL profile plan was first executed to warm the buffer cache.
4. Statistics for the SQL profile plan were averaged over
next 9 executions.
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information.
25-32
Chapter 25
Running SQL Tuning Advisor On Demand
25-33
Chapter 25
Running SQL Tuning Advisor On Demand
See Also:
Oracle Database 2 Day + Performance Tuning Guide to learn how to
configure and run SQL Tuning Advisor using Cloud Control.
To use the APIs, the user must have the ADVISOR privilege.
25-34
Chapter 25
Running SQL Tuning Advisor On Demand
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information
The following graphic shows the basic workflow when using the PL/SQL APIs.
STS
create_tuning_task
set_tuning_task_parameter
execute_tuning_task
Monitor Task
report_tuning_task
Implement
Recommendations
25-35
Chapter 25
Running SQL Tuning Advisor On Demand
See Also:
Oracle Database 2 Day + Performance Tuning Guide to learn how to tune
SQL using Cloud Control
Assumptions
This tutorial assumes the following:
• You want to tune as user hr, who has the ADVISOR privilege.
• You want to tune the following query:
25-36
Chapter 25
Running SQL Tuning Advisor On Demand
• You want to pass the bind variable 100 to the preceding query.
• You want SQL Tuning Advisor to perform SQL profiling.
• You want the task to run no longer than 60 seconds.
DECLARE
my_task_name VARCHAR2(30);
my_sqltext CLOB;
BEGIN
my_sqltext := 'SELECT /*+ ORDERED */ * ' ||
'FROM employees e, locations l, departments d ' ||
'WHERE e.department_id = d.department_id AND ' ||
'l.location_id = d.location_id AND ' ||
'e.employee_id < :bnd';
my_task_name := DBMS_SQLTUNE.CREATE_TUNING_TASK (
sql_text => my_sqltext
, bind_list => sql_binds(anydata.ConvertNumber(100))
, user_name => 'HR'
, scope => 'COMPREHENSIVE'
, time_limit => 60
, task_name => 'STA_SPECIFIC_EMP_TASK'
, description => 'Task to tune a query on a specified employee'
);
END;
/
25-37
Chapter 25
Running SQL Tuning Advisor On Demand
In the preceding output, the INITIAL status indicates that the task has not yet
started execution.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_SQLTUNE.CREATE_TUNING_TASK function
BEGIN
DBMS_SQLTUNE.SET_TUNING_TASK_PARAMETER (
task_name => 'STA_SPECIFIC_EMP_TASK'
, parameter => 'TIME_LIMIT'
, value => 300
);
END;
/
25-38
Chapter 25
Running SQL Tuning Advisor On Demand
PARAMETER_NAME VALUE
------------------------- ---------------
DAYS_TO_EXPIRE 30
DEFAULT_EXECUTION_TYPE TUNE SQL
EXECUTION_DAYS_TO_EXPIRE UNLIMITED
JOURNALING INFORMATION
MODE COMPREHENSIVE
SQL_LIMIT -1
SQL_PERCENTAGE 1
TARGET_OBJECTS 1
TEST_EXECUTE AUTO
TIME_LIMIT 300
The following program, which is issued on the standby database, shows a sample
SQL tuning session for a query of table1. The database_link_to parameter specifies
the name of the standby-to-primary database link.
BEGIN
:tname := DBMS_SQLTUNE.CREATE_TUNING_TASK(
sql_text => :query
, task_name => :tname
, database_link_to => 'lnk_to_pri' );
END;
/
EXEC DBMS_SQLTUNE.EXECUTE_TUNING_TASK(:tname);
BEGIN
DBMS_SQLTUNE.ACCEPT_SQL_PROFILE(
task_name => :tname
, name => 'prof'
, task_owner => 'SYS'
, replace => TRUE
, database_link_to => 'lnk_to_pri' );
25-39
Chapter 25
Running SQL Tuning Advisor On Demand
END;
/
See Also:
Note:
You can also execute the automatic tuning task SYS_AUTO_SQL_TUNING_TASK
using the EXECUTE_TUNING_TASK API. SQL Tuning Advisor performs the
same analysis and actions as it would when run automatically.
Assumptions
This tutorial assumes the following:
• You want to tune as user hr, who has the ADVISOR privilege.
• You want to execute the STA_SPECIFIC_EMP_TASK created in "Creating a SQL
Tuning Task".
BEGIN
DBMS_SQLTUNE.EXECUTE_TUNING_TASK(task_name=>'STA_SPECIFIC_EMP_TASK');
END;
/
25-40
Chapter 25
Running SQL Tuning Advisor On Demand
The following example queries the status of all tasks owned by the current user,
which in this example is hr:
See Also:
Oracle Database PL/SQL Packages and Types Reference for complete
reference information about the DBMS_SQLTUNE.EXECUTE_TUNING_TASK
function
View Description
USER_ADVISOR_TASKS Displays information about tasks owned by the current
user. The view contains one row for each task. Each
task has a name that is unique to the owner. Task
names are just informational and no uniqueness is
enforced within any other namespace.
V$ADVISOR_PROGRESS Displays information about the progress of advisor
execution.
Assumptions
This tutorial assumes the following:
• You tune as user hr, who has the ADVISOR privilege.
• You monitor the STA_SPECIFIC_EMP_TASK that you executed in "Executing a SQL
Tuning Task".
25-41
Chapter 25
Running SQL Tuning Advisor On Demand
SELECT STATUS
FROM USER_ADVISOR_TASKS
WHERE TASK_NAME = 'STA_SPECIFIC_EMP_TASK';
STATUS
-----------
EXECUTING
See Also:
Oracle Database Reference to learn about the V$ADVISOR_PROGRESS view
25-42
Chapter 25
Running SQL Tuning Advisor On Demand
Assumptions
This tutorial assumes the following:
• You want to tune as user hr, who has the ADVISOR privilege.
• You want to access the report for the STA_SPECIFIC_EMP_TASK executed in
"Executing a SQL Tuning Task".
DBMS_SQLTUNE.REPORT_TUNING_TASK('STA_SPECIFIC_EMP_TASK')
------------------------------------------------------------------------
---
GENERAL INFORMATION SECTION
------------------------------------------------------------------------
---
Tuning Task Name : STA_SPECIFIC_EMP_TASK
Tuning Task Owner : HR
Workload Type : Single SQL Statement
Execution Count : 11
Current Execution : EXEC_1057
Execution Type : TUNE SQL
Scope : COMPREHENSIVE
Time Limit(seconds): 300
Completion Status : COMPLETED
Started at : 04/22/2012 07:35:49
Completed at : 04/22/2012 07:35:50
------------------------------------------------------------------------
---
Schema Name: HR
SQL ID : dg7nfaj0bdcvk
SQL Text : SELECT /*+ ORDERED */ * FROM employees e, locations l,
departments d WHERE e.department_id = d.department_id AND
l.location_id = d.location_id AND e.employee_id < :bnd
Bind Variables :
1 - (NUMBER):100
------------------------------------------------------------------------
---
25-43
Chapter 25
Running SQL Tuning Advisor On Demand
See Also:
25-44
26
Optimizing Access Paths with SQL Access
Advisor
SQL Access Advisor is diagnostic software that identifies and helps resolve SQL
performance problems by recommending indexes, materialized views, materialized
view logs, or partitions to create, drop, or retain.
This chapter contains the following topics:
• About SQL Access Advisor
SQL Access Advisor accepts input from several sources, including SQL tuning
sets, and then issues recommendations.
• Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Basic tasks include creating an STS, loading it, creating a SQL Access Advisor
task, and then executing the task.
• Performing a SQL Access Advisor Quick Tune
To tune a single SQL statement, the DBMS_ADVISOR.QUICK_TUNE procedure
accepts as its input a task_name and a single SQL statement.
• Using SQL Access Advisor: Advanced Tasks
This section describes advanced tasks involving SQL Access Advisor.
• SQL Access Advisor Examples
Oracle Database provides a script that contains several SQL Access Advisor
examples that you can run on a test database.
• SQL Access Advisor Reference
You can access metadata about SQL Access Advisor using data dictionary views.
Note:
Data visibility and privilege requirements may differ when using SQL Access
Advisor with pluggable databases.
26-1
Chapter 26
About SQL Access Advisor
See Also:
Oracle Database Administrator’s Guide for a table that summarizes how
manageability features work in a container database (CDB)
See Also:
26-2
Chapter 26
About SQL Access Advisor
Filter Options
Hypothetical Optimizer
Automatic
Tuning
Shared Pool Optimizer
Library Cache
Materialized
Views
SQL Materialized
Tuning View Logs
Set
Partitions
DBA
See Also:
"About Automatic Tuning Optimizer"
26-3
Chapter 26
About SQL Access Advisor
Note:
For best results, provide a workload as a SQL tuning set. The
DBMS_SQLTUNE package provides helper functions that can create SQL
tuning sets from common workload sources, such as the SQL cache, a
user-defined workload stored in a table, and a hypothetical workload.
• Hypothetical workload
You can create a hypothetical workload from a schema by analyzing dimensions
and constraints. This option is useful when you are initially designing your
application.
See Also:
26-4
Chapter 26
About SQL Access Advisor
process, such as the type of recommendation required and the naming conventions for
what it recommends.
To set these parameters, use the DBMS_ADVISOR.SET_TASK_PARAMETER procedure.
Parameters are persistent in that they remain set for the life span of the task. When a
parameter value is set using DBMS_ADVISOR.SET_TASK_PARAMETER, the value does not
change until you make another call to this procedure.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_ADVISOR.SET_TASK_PARAMETER procedure
26-5
Chapter 26
About SQL Access Advisor
See Also:
26-6
Chapter 26
About SQL Access Advisor
• GATHER STATS
This action generates a call to a DBMS_STATS procedure to gather statistics on a
newly generated access structure.
Multiple recommendations may refer to the same action. However, when generating a
script for the recommendation, you only see each action once.
See Also:
26-7
Chapter 26
About SQL Access Advisor
See Also:
• Accessing the SQL Access Advisor: Initial Options Page Using Cloud Control
The SQL Access Advisor: Initial Options page in Cloud Control is the starting page
for a wizard that guides you through the process of obtaining recommendations.
• Command-Line Interface to SQL Tuning Sets
On the command line, you can use the DBMS_ADVISOR package to manage SQL
Tuning Advisor.
See Also:
26-8
Chapter 26
About SQL Access Advisor
26.1.3.1 Accessing the SQL Access Advisor: Initial Options Page Using Cloud
Control
The SQL Access Advisor: Initial Options page in Cloud Control is the starting page for
a wizard that guides you through the process of obtaining recommendations.
You can perform most SQL plan management tasks in this page or in pages
accessed through this page.
See Also:
26-9
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about
DBMS_ADVISOR
26-10
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Typically, you use SQL Access Advisor by performing the following steps:
1. Create a SQL tuning set
The input workload source for SQL Access Advisor is a SQL tuning set (STS). Use
DBMS_SQLTUNE.CREATE_SQLSET to create a SQL tuning set.
"Creating a SQL Tuning Set as Input for SQL Access Advisor" describes this task.
2. Load the SQL tuning set
SQL Access Advisor performs best when a workload based on actual usage is
available. Use DBMS_SQLTUNE.LOAD_SQLSET to populate the SQL tuning set with
your workload.
"Populating a SQL Tuning Set with a User-Defined Workload" describes this task.
3. Create and configure a task
In the task, you define what SQL Access Advisor must analyze and the location of
the analysis results. Create a task using the DBMS_ADVISOR.CREATE_TASK
procedure. You can then define parameters for the task using the
SET_TASK_PARAMETER procedure, and then link the task to an STS by using the
DBMS_ADVISOR.ADD_STS_REF procedure.
"Creating and Configuring a SQL Access Advisor Task" describes this task.
26-11
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
26.2.1 Creating a SQL Tuning Set as Input for SQL Access Advisor
The input workload source for SQL Access Advisor is an STS.
Because an STS is stored as a separate entity, multiple advisor tasks can share it.
Create an STS with the DBMS_SQLTUNE.CREATE_SQLSET procedure.
After an advisor task has referenced an STS, you cannot delete or modify the STS
until all advisor tasks have removed their dependency on it. A workload reference is
removed when a parent advisor task is deleted, or when you manually remove the
workload reference from the advisor task.
26-12
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Prerequisites
The user creating the STS must have been granted the ADMINISTER SQL TUNING SET
privilege. To run SQL Access Advisor on SQL tuning sets owned by other users, the
user must have the ADMINISTER ANY SQL TUNING SET privilege.
Assumptions
This tutorial assumes the following:
• You want to create an STS named MY_STS_WORKLOAD.
• You want to use this STS as input for a workload derived from the sh schema.
To create an STS :
1. In SQL*Plus, log in to the database as user sh.
2. Set SQL*Plus variables.
For example, enter the following commands:
See Also:
26-13
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
You cannot use SQL Access Advisor without a workload. SQL Access Advisor ranks
the entries according to a specific statistic, business importance, or combination of the
two, which enables the advisor to process the most important SQL statements first.
SQL Access Advisor performs best with a workload based on actual usage. You can
store multiple workloads in the form of SQL tuning sets, so that you can view the
different uses of a real-world data warehousing or OLTP environment over a long
period and across the life cycle of database instance startup and shutdown.
The following table describes procedures that you can use to populate an STS with a
user-defined workload.
Assumptions
This tutorial assumes that you want to do the following:
• Create a table named sh.user_workload to store information about SQL
statements
• Load the sh.user_workload table with information about three queries of tables in
the sh schema
• Populate the STS created in "Creating a SQL Tuning Set as Input for SQL Access
Advisor" with the workload contained in sh.user_workload
26-14
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
*/
elapsed_time number, /* Elapsed time for query
*/
cpu_time number, /* CPU time for query
*/
buffer_gets number, /* Buffer gets consumed by query
*/
disk_reads number, /* Disk reads consumed by query
*/
rows_processed number, /* # of rows processed by query
*/
executions number, /* # of times query executed
*/
optimizer_cost number, /* Optimizer cost for query
*/
priority number, /* User-priority (1,2 or 3)
*/
last_execution_date date, /* Last time query executed
*/
stat_period number, /* Window exec time in seconds
*/
sql_text clob /* Full SQL Text
*/
);
-- order by
26-15
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
3. Execute a PL/SQL program that fills a cursor with rows from the user_workload
table, and then loads the contents of this cursor into the STS named
MY_STS_WORKLOAD.
For example, execute the following PL/SQL program:
DECLARE
sqlset_cur DBMS_SQLTUNE.SQLSET_CURSOR;
BEGIN
OPEN sqlset_cur FOR
SELECT SQLSET_ROW(null,null, SQL_TEXT, null, null, 'SH', module,
'Action', 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, null, 2, 3,
sysdate, 0, 0, null, 0, null, null)
FROM USER_WORKLOAD;
DBMS_SQLTUNE.LOAD_SQLSET('MY_STS_WORKLOAD', sqlset_cur);
END;
/
In the SQL Access Advisor task, you define what the advisor must analyze and the
location of the results. You can create multiple tasks, each with its own specialization.
All are based on the same Advisor task model and share the same repository.
Configuring the task involves the following steps:
• Defining task parameters
At the time the recommendations are generated, you can apply a filter to the
workload to restrict what is analyzed. This restriction provides the ability to
generate different sets of recommendations based on different workload
scenarios.
SQL Access Advisor parameters control the recommendation process and
customization of the workload. These parameters control various aspects of the
process, such as the type of recommendation required and the naming
conventions for what it recommends.
If parameters are not defined, then the database uses the defaults. You can set
task parameters by using the DBMS_ADVISOR.SET_TASK_PARAMETER procedure.
Parameters are persistent in that they remain set for the life span of the task.
When a parameter value is set using SET_TASK_PARAMETER, it does not change
until you make another call to this procedure.
• Linking the task to the workload
26-16
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Because the workload is independent, you must link it to a task using the
DBMS_ADVISOR.ADD_STS_REF procedure. After this link has been established, you
cannot delete or modify the workload until all advisor tasks have removed their
dependency on the workload. A workload reference is removed when a user
deletes a parent advisor task or manually removes the workload reference from
the task by using the DBMS_ADVISOR.DELETE_STS_REF procedure.
Prerequisites
The user creating the task must have been granted the ADVISOR privilege.
Assumptions
This tutorial assumes the following:
• You want to create a task named MYTASK.
• You want to use this task to analyze the workload that you defined in "Populating a
SQL Tuning Set with a User-Defined Workload".
• You want to terminate the task if it takes longer than 30 minutes to execute.
• You want to SQL Access Advisor to only consider indexes.
26-17
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
See Also:
Prerequisites
When processing a workload, SQL Access Advisor attempts to validate each
statement to identify table and column references. The database achieves validation
by processing each statement as if it were being executed by the statement's original
user.
If the user does not have SELECT privileges to a particular table, then SQL Access
Advisor bypasses the statement referencing the table. This behavior can cause many
statements to be excluded from analysis. If SQL Access Advisor excludes all
statements in a workload, then the workload is invalid. SQL Access Advisor returns the
following message:
To avoid missing critical workload queries, the current database user must have
SELECT privileges on the tables targeted for materialized view analysis. For these
tables, these SELECT privileges cannot be obtained through a role.
Assumptions
This tutorial assumes that you want to execute the task you configured in "Creating
and Configuring a SQL Access Advisor Task".
26-18
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
EXECUTE DBMS_ADVISOR.EXECUTE_TASK(:task_name);
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the EXECUTE_TASK procedure and its parameters
26-19
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Assumptions
This tutorial assumes that you want to view results of the task you executed in
"Executing a SQL Access Advisor Task".
26-20
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
1 1 236
2 2 356
The precost and postcost numbers are in terms of the estimated optimizer cost
(shown in EXPLAIN PLAN) both without and with the recommended access structure
changes.
3. Display the number of distinct actions for this set of recommendations.
For example, use the following query (sample output included):
'ACTIONCOUNT CNT
------------ ----------
Action Count 4
26-21
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
2 1 PARTITION TABLE
2 3 RETAIN INDEX
2 4 RETAIN INDEX
=========================================
Task_name = MYTASK
Action ID: 1
Command : PARTITION TABLE
Attr1 (name) : "SH"."SALES"
Attr2 (tablespace):
26-22
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
Attr3 : ("TIME_ID")
Attr4 : INTERVAL
Attr5 :
----------------------------------------
Action ID: 2
Command : RETAIN INDEX
Attr1 (name) : "SH"."PRODUCTS_PK"
Attr2 (tablespace):
Attr3 : "SH"."PRODUCTS"
Attr4 : BTREE
Attr5 :
----------------------------------------
Action ID: 3
Command : RETAIN INDEX
Attr1 (name) : "SH"."TIMES_PK"
Attr2 (tablespace):
Attr3 : "SH"."TIMES"
Attr4 : BTREE
Attr5 :
----------------------------------------
Action ID: 4
Command : RETAIN INDEX
Attr1 (name) : "SH"."SALES_TIME_BIX"
Attr2 (tablespace):
Attr3 : "SH"."SALES"
Attr4 : BITMAP
Attr5 :
----------------------------------------
=========END RECOMMENDATIONS============
See Also:
For new objects, the names of the materialized views, materialized view logs, and
indexes are automatically generated by using the user-specified name template.
Review the generated SQL script before attempting to execute it.
Assumptions
This tutorial assumes that you want to save and execute a script that contains the
recommendations generated in "Executing a SQL Access Advisor Task".
26-23
Chapter 26
Optimizing Access Paths with SQL Access Advisor: Basic Tasks
3. Connect to the database as sh, and then save the script to a file.
For example, use the following statement:
EXECUTE DBMS_ADVISOR.CREATE_FILE(DBMS_ADVISOR.GET_TASK_SCRIPT('MYTASK'),
'ADVISOR_RESULTS', 'advscript.sql');
Rem Username: SH
Rem Task: MYTASK
Rem Execution date:
Rem
Rem
Rem Repartitioning table "SH"."SALES"
Rem
SET SERVEROUTPUT ON
SET ECHO ON
Rem
Rem Creating new partitioned table
Rem
CREATE TABLE "SH"."SALES1"
( "PROD_ID" NUMBER,
"CUST_ID" NUMBER,
"TIME_ID" DATE,
"CHANNEL_ID" NUMBER,
"PROMO_ID" NUMBER,
"QUANTITY_SOLD" NUMBER(10,2),
"AMOUNT_SOLD" NUMBER(10,2)
) PCTFREE 5 PCTUSED 40 INITRANS 1 MAXTRANS 255
NOCOMPRESS NOLOGGING
TABLESPACE "EXAMPLE"
PARTITION BY RANGE ("TIME_ID") INTERVAL( NUMTOYMINTERVAL( 1, 'MONTH'))
( PARTITION VALUES LESS THAN (TO_DATE(' 1998-02-01 00:00:00',
'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN')) );
.
.
.
26-24
Chapter 26
Performing a SQL Access Advisor Quick Tune
@/tmp/advscript.sql
See Also:
Assumptions
This tutorial assumes the following:
• You want to tune a single SQL statement.
• You want to name the task MY_QUICKTUNE_TASK.
EXEC
DBMS_ADVISOR.QUICK_TUNE(DBMS_ADVISOR.SQLACCESS_ADVISOR,:t_name,:sq);
26-25
Chapter 26
Using SQL Access Advisor: Advanced Tasks
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the QUICK_TUNE procedure and its parameters
26-26
Chapter 26
Using SQL Access Advisor: Advanced Tasks
EXECUTE DBMS_ADVISOR.EXECUTE_TASK(:t_name);
EXECUTE
DBMS_ADVISOR.SET_TASK_PARAMETER(:t_name,'ANALYSIS_SCOPE','EVALUATION');
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
SET_TASK_PARAMETER procedure and its parameters
Assumptions
This tutorial assumes the following:
• You want to change the name of existing task MYTASK to TUNING1.
• You want to make the task TUNING1 read-only.
26-27
Chapter 26
Using SQL Access Advisor: Advanced Tasks
EXECUTE DBMS_ADVISOR.UPDATE_TASK_ATTRIBUTES('TUNING1',
read_only => 'true');
See Also:
The following table describes procedures that you can use to manage task templates.
Procedure Description
CREATE_TASK The template parameter is an optional task name of an existing task or
task template. To specify built-in SQL Access Advisor templates, use the
template name as described in Table 26-6. is_template is an optional
parameter that enables you to set the newly created task as a template.
Valid values are true and false.
SET_TASK_PARA The INDEX_NAME_TEMPLATE parameter specifies the method by which new
METER index names are formed. The MVIEW_NAME_TEMPLATE parameter specifies
the method by which new materialized view names are formed. The
PARTITION_NAME_TEMPLATE parameter specifies the method by which
new partition names are formed.
UPDATE_TASK_A is_template marks the task as a template. Physically, there is no
TTRIBUTES difference between a task and a template; however, a template cannot be
executed. Possible values are: true and false. If the value is NULL or
contains the value ADVISOR_UNUSED, then the setting is not changed.
26-28
Chapter 26
Using SQL Access Advisor: Advanced Tasks
Assumptions
This tutorial assumes the following:
• You want to create a template named MY_TEMPLATE.
• You want to set naming conventions for indexes and materialized views that are
recommended by tasks based on MY_TEMPLATE.
• You want to create task NEWTASK based on MY_TEMPLATE.
BEGIN
DBMS_ADVISOR.SET_TASK_PARAMETER (
:template_name
, 'MVIEW_NAME_TEMPLATE'
, 'SH_MV$$_<SEQ>'
);
END;
26-29
Chapter 26
Using SQL Access Advisor: Advanced Tasks
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
CREATE_TASK and SET_TASK_PARAMETER procedures
26-30
Chapter 26
Using SQL Access Advisor: Advanced Tasks
Assumptions
This tutorial assumes the following:
• Long-running task MYTASK is currently executing.
• You want to interrupt this task, and then view the recommendations.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
INTERRUPT_TASK procedure
26-31
Chapter 26
Using SQL Access Advisor: Advanced Tasks
The RESET_TASK procedure resets a task to its initial starting point, which has the effect
of removing all recommendations and intermediate data from the task. The task status
is set to INITIAL. The syntax is as follows:
Assumptions
This tutorial assumes the following:
• Long-running task MYTASK is currently executing. This task is set to make
partitioning recommendations.
• You want to cancel this task, and then reset it so that the task makes only index
recommendations.
EXECUTE DBMS_ADVISOR.RESET_TASK('MYTASK');
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about RESET_TASK and CANCEL_TASK
26-32
Chapter 26
Using SQL Access Advisor: Advanced Tasks
If a task is linked to an STS workload, and if you want to delete the task or workload,
then you must remove the link between the task and the workload using the
DELETE_STS_REF procedure. The following example deletes the link between task
MYTASK and the current user's SQL tuning set MY_STS_WORKLOAD:
Assumptions
This tutorial assumes the following:
• User sh currently owns multiple SQL Access Advisor tasks.
• You want to delete MYTASK.
• The task MYTASK is currently linked to workload MY_STS_WORKLOAD.
SELECT TASK_NAME
FROM USER_ADVISOR_TASKS
WHERE ADVISOR_NAME = 'SQL Access Advisor';
TASK_NAME
-------------------------
MYTASK
NEWTASK
EXECUTE DBMS_ADVISOR.DELETE_TASK('MYTASK');
26-33
Chapter 26
Using SQL Access Advisor: Advanced Tasks
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DELETE_TASK procedure and its parameters
DBMS_ADVISOR.MARK_RECOMMENDATION (
task_name IN VARCHAR2
id IN NUMBER,
action IN VARCHAR2);
Assumptions
This tutorial assumes the following:
• You are reviewing the recommendations as described in tutorial "Viewing SQL
Access Advisor Task Results".
• You want to reject the first recommendation, which partitions a table.
To mark a recommendation:
1. Connect SQL*Plus to the database as user sh, and then mark the
recommendation.
For example, reject recommendation 1 as follows:
26-34
Chapter 26
Using SQL Access Advisor: Advanced Tasks
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the MARK_RECOMMENDATIONS procedure and its parameters
DBMS_ADVISOR.UPDATE_REC_ATTRIBUTES (
task_name IN VARCHAR2
rec_id IN NUMBER,
action_id IN NUMBER,
attribute_name IN VARCHAR2,
value IN VARCHAR2);
• OWNER
Specifies the owner name of the recommended object.
• NAME
Specifies the name of the recommended object.
• TABLESPACE
Specifies the tablespace of the recommended object.
Assumptions
This tutorial assumes the following:
• You are reviewing the recommendations as described in tutorial "Viewing SQL
Access Advisor Task Results".
• You want to change the tablespace for recommendation 1, action 1 to SH_MVIEWS.
To mark a recommendation:
1. Connect SQL*Plus to the database as user sh, and then update the
recommendation attribute.
For example, change the tablespace name to SH_MVIEWS as follows:
BEGIN
DBMS_ADVISOR.UPDATE_REC_ATTRIBUTES (
'MYTASK'
26-35
Chapter 26
SQL Access Advisor Examples
, 1
, 1
, 'TABLESPACE'
, 'SH_MVIEWS'
);
END;
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the UPDATE_REC_ATTRIBUTES procedure and its parameters
The following table maps SQL Access Advisor actions to attribute columns in the
DBA_ADVISOR_ACTIONS view. In the table, MV refers to a materialized view.
26-36
Chapter 26
SQL Access Advisor Reference
26-37
Chapter 26
SQL Access Advisor Reference
26-38
Chapter 26
SQL Access Advisor Reference
Table 26-5 (Cont.) Types of Advisor Task Parameters And Their Uses
You can use the constants shown in the following table with SQL Access Advisor.
Constant Description
ADVISOR_ALL A value that indicates all possible values. For string parameters,
this value is equivalent to the wildcard (%) character.
ADVISOR_CURRENT Indicates the current time or active set of elements. Typically, this
is used in time parameters.
ADVISOR_DEFAULT Indicates the default value. Typically used when setting task or
workload parameters.
ADVISOR_UNLIMITED A value that represents an unlimited numeric value.
ADVISOR_UNUSED A value that represents an unused entity. When a parameter is
set to ADVISOR_UNUSED, it has no effect on the current operation.
A typical use for this constant is to set a parameter as unused for
its dependent operations.
SQLACCESS_GENERAL Specifies the name of a default SQL Access general-purpose task
template. This template sets the DML_VOLATILITY task
parameter to true and ANALYSIS_SCOPE to INDEX, MVIEW.
SQLACCESS_OLTP Specifies the name of a default SQL Access OLTP task template.
This template sets the DML_VOLATILITY task parameter to true
and ANALYSIS_SCOPE to INDEX.
SQLACCESS_WAREHOUSE Specifies the name of a default SQL Access warehouse task
template. This template sets the DML_VOLATILITY task
parameter to false and EXECUTION_TYPE to INDEX, MVIEW.
SQLACCESS_ADVISOR Contains the formal name of SQL Access Advisor. You can
specify this name when procedures require the Advisor name as
an argument.
26-39
Part IX
SQL Management Objects
A SQL management object is a feature that stabilizes the execution plans of
individual SQL statements. SQL profiles and SQL plan baselines are SQL
management objects.
This part contains the following chapters:
• Managing SQL Profiles
When warranted, SQL Tuning Advisor recommends a SQL profile. You can use
DBMS_SQLTUNE to implement, alter, drop, and transport SQL profiles.
• Overview of SQL Plan Management
SQL plan management is a preventative mechanism that enables the optimizer
to automatically manage execution plans, ensuring that the database uses only
known or verified plans.
• Managing SQL Plan Baselines
This chapter explains the concepts and tasks relating to SQL plan management
using the DBMS_SPM package.
• Migrating Stored Outlines to SQL Plan Baselines
Stored outline migration is the user-initiated process of converting stored
outlines to SQL plan baselines. A SQL plan baseline is a set of plans proven to
provide optimal performance.
27
Managing SQL Profiles
When warranted, SQL Tuning Advisor recommends a SQL profile. You can use
DBMS_SQLTUNE to implement, alter, drop, and transport SQL profiles.
27-1
Chapter 27
About SQL Profiles
See Also:
"About SQL Tuning Advisor"
See Also:
Note:
The SQL profile contains supplemental statistics for the entire statement, not
individual plans. The profile does not itself determine a specific plan.
27-2
Chapter 27
About SQL Profiles
A SQL profile contains, among other statistics, a set of cardinality adjustments. The
cardinality measure is based on sampling the WHERE clause rather than on statistical
projection. A profile uses parts of the query to determine whether the estimated
cardinalities are close to the actual cardinalities and, if a mismatch exists, uses the
corrected cardinalities. For example, if a SQL profile exists for SELECT * FROM t
WHERE x=5 AND y=10, then the profile stores the actual number of rows returned.
When choosing plans, the optimizer has the following sources of information:
• The environment, which contains the database configuration, bind variable values,
optimizer statistics, data set, and so on
• The supplemental statistics in the SQL profile
The following figure shows the relationship between a SQL statement and the SQL
profile for this statement. The optimizer uses the SQL profile and the environment to
generate an execution plan. In this example, the plan is in the SQL plan baseline for
the statement.
SELECT . . . GB
HJ
HJ
Environment GB
Optimizer
Statistics Data NL
Set NL
Bind Configuration
Variables
If either the optimizer environment or SQL profile changes, then the optimizer can
create a new plan. As tables grow, or as indexes are created or dropped, the plan for a
SQL profile can change. The profile continues to be relevant even if the data
distribution or access path of the corresponding statement changes.
In general, you do not need to refresh SQL profiles. Over time, however, profile
content can become outdated. In this case, performance of the SQL statement may
degrade. The statement may appear as high-load or top SQL. In this case, the
Automatic SQL Tuning task again captures the statement as high-load SQL. You can
implement a new SQL profile for the statement.
Internally, a SQL profile is implemented using hints that address different types of
problems. These hints do not specify any particular plan. Rather, the hints correct
errors in the optimizer estimation algorithm that lead to suboptimal plans. For example,
a profile may use the TABLE_STATS hint to set object statistics for tables when the
statistics are missing or stale.
This section contains the following topics:
27-3
Chapter 27
About SQL Profiles
See Also:
---------------------------------------------------------------------------
----
FINDINGS SECTION (2 findings)
---------------------------------------------------------------------------
----
Validation results
------------------
The SQL profile was tested by executing both its plan and the original
plan
and measuring their respective execution statistics. A plan may have been
27-4
Chapter 27
About SQL Profiles
Notes
-----
1. The SQL profile plan was first executed to warm the buffer cache.
2. Statistics for the SQL profile plan were averaged over next 3
executions.
Sometimes SQL Tuning Advisor may recommend implementing a profile that uses the
Automatic Degree of Parallelism (Auto DOP) feature. A parallel query profile is only
recommended when the original plan is serial and when parallel execution can
significantly reduce the elapsed time for a long-running query.
When it recommends a profile that uses Auto DOP, SQL Tuning Advisor gives details
about the performance overhead of using parallel execution for the SQL statement in
the report. For parallel execution recommendations, SQL Tuning Advisor may provide
two SQL profile recommendations, one using serial execution and one using parallel.
The following example shows a parallel query recommendation. In this example, a
degree of parallelism of 7 improves response time significantly at the cost of
increasing resource consumption by almost 25%. You must decide whether the
reduction in database throughput is worth the increase in response time.
Executing this query parallel with DOP 7 will improve its response time
82.22% over the SQL profile plan. However, there is some cost in enabling
parallel execution. It will increase the statement's resource
consumption by
an estimated 24.43% which may result in a reduction of system throughput.
Also, because these resources are consumed over a much smaller duration,
the
response time of concurrent statements might be negatively impacted if
sufficient hardware capacity is not available.
27-5
Chapter 27
About SQL Profiles
The following data shows some sampled statistics for this SQL from the
past
week and projected weekly values when parallel execution is enabled.
See Also:
• "SQL Profiling"
• Oracle Database VLDB and Partitioning Guide to learn more about Auto
DOP
• Oracle Database PL/SQL Packages and Types Reference to learn about
the DBMS_SQLTUNE.ACCEPT_SQL_PROFILE procedure
See Also:
"Overview of SQL Plan Management"
27-6
Chapter 27
About SQL Profiles
See Also:
DBMS_SQLTUNE. Implement a
ACCEPT_SQL_PROFILE SQL Profile
DBMS_SQLTUNE.
DBA_SQL_PROFILES
ALTER_SQL_PROFILE
27-7
Chapter 27
Implementing a SQL Profile
See Also:
Oracle Database PL/SQL Packages and Types Reference for information
about the DBMS_SQLTUNE package
27-8
Chapter 27
Implementing a SQL Profile
See Also:
Oracle Database PL/SQL Packages and Types Reference for information
about the ACCEPT_SQL_PROFILE procedure
Assumptions
This tutorial assumes the following:
• The SQL Tuning Advisor task STA_SPECIFIC_EMP_TASK includes a
recommendation to create a SQL profile.
• The name of the SQL profile is my_sql_profile.
• The PL/SQL block accepts a profile that uses parallel execution (profile_type).
• The profile uses force matching.
DECLARE
my_sqlprofile_name VARCHAR2(30);
BEGIN
my_sqlprofile_name := DBMS_SQLTUNE.ACCEPT_SQL_PROFILE (
task_name => 'STA_SPECIFIC_EMP_TASK'
, name => 'my_sql_profile'
, profile_type => DBMS_SQLTUNE.PX_PROFILE
, force_match => true
);
27-9
Chapter 27
Listing SQL Profiles
END;
/
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DBMS_SQLTUNE.ACCEPT_SQL_PROFILE procedure
See Also:
Oracle Database Reference to learn about the DBA_SQL_PROFILES view
27-10
Chapter 27
Altering a SQL Profile
The CATEGORY attribute determines which sessions can apply a profile. View the
CATEGORY attribute by querying DBA_SQL_PROFILES.CATEGORY. By default, all profiles
are in the DEFAULT category, which means that all sessions in which the
SQLTUNE_CATEGORY initialization parameter is set to DEFAULT can use the profile.
By altering the category of a SQL profile, you determine which sessions are affected
by profile creation. For example, by setting the category to DEV, only sessions in which
the SQLTUNE_CATEGORY initialization parameter is set to DEV can use the profile. Other
sessions do not have access to the SQL profile and execution plans for SQL
statements are not impacted by the SQL profile. This technique enables you to test a
profile in a restricted environment before making it available to other sessions.
The example in this section assumes that you want to change the category of the SQL
profile so it is used only by sessions with the SQL profile category set to TEST, run the
SQL statement, and then change the profile category back to DEFAULT.
27-11
Chapter 27
Dropping a SQL Profile
);
END;
See Also:
Assumptions
This section assumes the following:
• You want to drop my_sql_profile.
• You want to ignore errors raised if the name does not exist.
BEGIN
DBMS_SQLTUNE.DROP_SQL_PROFILE (
name => 'my_sql_profile'
);
END;
/
See Also:
27-12
Chapter 27
Transporting a SQL Profile
The following graphic shows the basic workflow of transporting SQL profiles.
DBMS_SQLTUNE Accept a
ACCEPT_SQL_PROFILE SQL Profile
CREATE_STGTAB_SQLPROF
PACK_STGTAB_SQLPROF
Transport
Transport SQL Profile to different
database
UNPACK_STGTAB_SQLPROF
Assumptions
This tutorial assumes the following:
• You want to transport my_profile from a production database to a test database.
• You want to create the staging table in the dba1 schema.
27-13
Chapter 27
Transporting a SQL Profile
BEGIN
DBMS_SQLTUNE.CREATE_STGTAB_SQLPROF (
table_name => 'my_staging_table'
, schema_name => 'dba1'
);
END;
/
2. Use the PACK_STGTAB_SQLPROF procedure to export SQL profiles into the staging
table.
The following example populates dba1.my_staging_table with the SQL profile
my_profile:
BEGIN
DBMS_SQLTUNE.PACK_STGTAB_SQLPROF (
profile_name => 'my_profile'
, staging_table_name => 'my_staging_table'
, staging_schema_owner => 'dba1'
);
END;
/
3. Move the staging table to the database where you plan to unpack the SQL
profiles.
Move the table using your utility of choice. For example, use Oracle Data Pump or
a database link.
4. On the database where you plan to import the SQL profiles, use
UNPACK_STGTAB_SQLPROF to unpack SQL profiles from the staging table.
The following example shows how to unpack SQL profiles in the staging table:
BEGIN
DBMS_SQLTUNE.UNPACK_STGTAB_SQLPROF(
replace => true
, staging_table_name => 'my_staging_table'
);
END;
/
See Also:
27-14
28
Overview of SQL Plan Management
SQL plan management is a preventative mechanism that enables the optimizer to
automatically manage execution plans, ensuring that the database uses only known or
verified plans.
SQL plan management uses a mechanism called a SQL plan baseline. A plan
baseline is a set of accepted plans that the optimizer is allowed to use for a SQL
statement. In this context, a plan includes all plan-related information (for example,
SQL plan identifier, set of hints, bind values, and optimizer environment) that the
optimizer needs to reproduce an execution plan. In the typical use case, the database
accepts a plan into the plan baseline only after verifying that the plan performs well.
The main components of SQL plan management are as follows:
• Plan capture
This component stores relevant information about plans for a set of SQL
statements.
• Plan selection
This component is the detection by the optimizer of plan changes based on stored
plan history, and the use of SQL plan baselines to select appropriate plans to
avoid potential performance regressions.
• Plan evolution
This component is the process of adding new plans to existing SQL plan
baselines, either manually or automatically.
This section contains the following topics:
• Purpose of SQL Plan Management
• Plan Capture
• Plan Selection
• Plan Evolution
• Storage Architecture for SQL Plan Management
• Purpose of SQL Plan Management
SQL plan management prevents performance regressions caused by plan
changes.
• Plan Capture
SQL plan capture refers to techniques for capturing and storing relevant
information about plans in the SQL Management Base for a set of SQL
statements.
• Plan Selection
SQL plan selection is the optimizer ability to detect plan changes based on
stored plan history, and the use of SQL plan baselines to select plans to avoid
potential performance regressions.
28-1
Chapter 28
Purpose of SQL Plan Management
• Plan Evolution
In general, SQL plan evolution is the process by which the optimizer verifies new
plans and adds them to an existing SQL plan baseline.
• Storage Architecture for SQL Plan Management
The SQL plan management infrastructure records the signatures of parsed
statements, and both accepted and unaccepted plans.
Note:
SQL plan baselines cannot help when an event has caused irreversible
execution plan changes, such as dropping an index.
28-2
Chapter 28
Purpose of SQL Plan Management
• Deployment of new application modules introduces new SQL statements into the
database.
The application software may use appropriate SQL execution plans developed in a
standard test configuration for the new statements. If the system configuration is
significantly different from the test configuration, then the database can evolve
SQL plan baselines over time to produce better performance.
See Also:
Oracle Database Upgrade Guide to learn how to upgrade an Oracle
database
Suboptimal Corrects
SQL Profile
Plan Cause
• SQL plan baselines reproduce a specific plan, whereas SQL profiles correct
optimizer cost estimates.
A SQL plan baseline is a set of accepted plans. Each plan is implemented using a
set of outline hints that fully specify a particular plan. SQL profiles are also
implemented using hints, but these hints do not specify any specific plan. Rather,
the hints correct miscalculations in the optimizer estimates that lead to suboptimal
plans. For example, a hint may correct the cardinality estimate of a table.
28-3
Chapter 28
Plan Capture
Because a profile does not constrain the optimizer to any one plan, a SQL profile
is more flexible than a SQL plan baseline. For example, changes in initialization
parameters and optimizer statistics enable the optimizer to choose a better plan.
Oracle recommends that you use SQL Tuning Advisor. In this way, you follow the
recommendations made by the advisor for SQL profiles and plan baselines rather than
trying to determine which mechanism is best for each SQL statement.
See Also:
28-4
Chapter 28
Plan Capture
See Also:
• "Plan Selection"
• Oracle Database Reference to learn about the
OPTIMIZER_USE_SQL_PLAN_BASELINES initialization parameter
28-5
Chapter 28
Plan Capture
See Also:
28-6
Chapter 28
Plan Capture
Shared Pool
Library Cache
SELECT * FROM
employees
SQL Management Base
GB
accepted
enabled
Staging Table HJ
HJ
Stored Outline
/*+ hint */
/*+ hint */
/*+ hint */
The loading behavior varies depending on whether a SQL plan baseline exists for
each statement represented in the bulk load:
• If a baseline for the statement does not exist, then the database does the
following:
1. Creates a plan history and plan baseline for the statement
2. Marks the initial plan for the statement as accepted
3. Adds the plan to the new baseline
• If a baseline for the statement exists, then the database does the following:
1. Marks the loaded plan as accepted
2. Adds the plan to the plan baseline for the statement without verifying the
plan's performance
Manually loaded plans are always marked accepted because the optimizer assumes
that any plan loaded manually by the administrator has acceptable performance. You
can load plans without enabling them by setting the enabled parameter to NO in the
DBMS_SPM.LOAD_PLANS_FROM_% functions.
28-7
Chapter 28
Plan Selection
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DBMS_SPM.LOAD_PLANS_FROM_% functions
28-8
Chapter 28
Plan Evolution
SQL is issued
Generate execution
plan
No No
Compare costs of
accepted plans
Execute lowest-cost
plan in baseline
See Also:
"Fixed Plans"
28-9
Chapter 28
Plan Evolution
Verifying Adding
28-10
Chapter 28
Storage Architecture for SQL Plan Management
See Also:
28-11
Chapter 28
Storage Architecture for SQL Plan Management
SYSAUX
SQL
Statement SQL
Log Profiles
SQL SQL
Plan
History Patches
Note:
Data visibility and privilege requirements may differ when using the SMB with
pluggable databases. See Oracle Database Administrator’s Guide for a table
that summarizes how manageability features work in a container database
(CDB).
See Also:
Oracle Database Administrator’s Guide to learn about the SYSAUX tablespace
Note:
If a filter excludes a statement, then its signature is also excluded from the
log.
28-12
Chapter 28
Storage Architecture for SQL Plan Management
System altered.
no rows selected
JOB_TITLE
-----------------------------------
President
SIGNATURE BATCH#
---------- ----------
1.8096E+19 1
Now the session executes a different jobs query. The log shows two tracked
statements:
JOB_TITLE
-----------------------------------
Public Relations Representative
SIGNATURE BATCH#
---------- ----------
1.7971E+19 1
1.8096E+19 1
no rows selected
28-13
Chapter 28
Storage Architecture for SQL Plan Management
The session executes the query for job_id='PR_REP' a second time. Because this
statement is now repeatable, and because automatic SQL plan capture is enabled, the
database creates a plan baseline for this statement. The query for job_id='AD_PRES'
has only been executed once, so no plan baseline exists for it.
JOB_TITLE
-----------------------------------
Public Relations Representative
SQL_HANDLE SQL_TEXT
-------------------- --------------------
SQL_f9676a330f972dd5 SELECT job_title FRO
M hr.jobs WHERE job_
id='PR_REP'
See Also:
• Enabled Plans
An enabled plan is a plan that is eligible for use by the optimizer.
28-14
Chapter 28
Storage Architecture for SQL Plan Management
• Accepted Plans
An accepted plan is a plan that is in a SQL plan baseline for a SQL statement
and thus available for use by the optimizer. An accepted plan contains a set of
hints, a plan hash value, and other plan-related information.
• Fixed Plans
A fixed plan is an accepted plan that is marked as preferred, so that the optimizer
considers only the fixed plans in the baseline. Fixed plans influence the plan
selection process of the optimizer.
See Also:
28-15
Chapter 28
Storage Architecture for SQL Plan Management
GB fixed
accepted
DBA marks enabled
as fixed HJ Optimizer
HJ Considers
GB fixed
accepted
enabled
HJ
HJ
Ignores unless
fixed plans are
GB not reproducible
accepted
enabled
HJ
HJ
If new plans are added to a baseline that contains at least one enabled fixed plan, then
the optimizer cannot use the new plans until you manually declare them as fixed.
28-16
29
Managing SQL Plan Baselines
This chapter explains the concepts and tasks relating to SQL plan management using
the DBMS_SPM package.
See Also:
29-1
Chapter 29
About Managing SQL Plan Baselines
29-2
Chapter 29
About Managing SQL Plan Baselines
You can perform most SQL plan management tasks in this page or in pages
accessed through this page.
See Also:
29-3
Chapter 29
About Managing SQL Plan Baselines
See Also:
29-4
Chapter 29
Configuring SQL Plan Management
29-5
Chapter 29
Configuring SQL Plan Management
statements are eligible for plan capture. By default, no filters are configured, which
means that all repeatable statements are eligible for plan capture.
• OPTIMIZER_USE_SQL_PLAN_BASELINES=true
For any SQL statement that has an existing SQL plan baseline, the database
automatically adds new plans to the SQL plan baseline as unaccepted plans.
Note:
The settings of the preceding parameters are independent of each other. For
example, if OPTIMIZER_CAPTURE_SQL_PLAN_BASELINES is true, then the
database creates initial plan baselines for new statements even if
OPTIMIZER_USE_SQL_PLAN_BASELINES is false.
If the default behavior is what you intend, then skip this section.
The following sections explain how to change the default parameter settings from the
command line. If you use Cloud Control, then set these parameters in the SQL Plan
Baseline subpage. This section contains the following topics:
See Also:
• "Figure 29-1"
• "Automatic Initial Plan Capture"
• "Plan Selection"
29.2.1.1 Enabling Automatic Initial Plan Capture for SQL Plan Management
Setting the OPTIMIZER_CAPTURE_SQL_PLAN_BASELINES initialization parameter to true is
all that is necessary for the database to automatically create an initial SQL plan
baseline for any eligible SQL statement not already in the plan history.
By default, the database considers all repeatable SQL statements as eligible for
capture, with the following exceptions:
29-6
Chapter 29
Configuring SQL Plan Management
Caution:
By default, when automatic baseline capture is enabled, the database
creates a SQL plan baseline for every eligible repeatable statement,
including all recursive SQL and monitoring SQL. Thus, automatic capture
may result in the creation of an extremely large number of plan baselines. To
limit the statements that are eligible for plan baselines, configure filters using
the DBMS_SPM.CONFIGURE procedure.
The following sample output shows that automatic initial plan capture is disabled:
If the parameters are set as you intend, then skip the remaining steps.
3. To enable the automatic recognition of repeatable SQL statements and the
generation of SQL plan baselines for these statements, enter the following
statement:
See Also:
Oracle Database Reference to learn more about
OPTIMIZER_CAPTURE_SQL_PLAN_BASELINES
29-7
Chapter 29
Configuring SQL Plan Management
Parameter Description
parameter_name The type of filter for automatic capture.
Possible values are AUTO_CAPTURE_SQL_TEXT,
AUTO_CAPTURE_PARSING_SCHEMA_NAME, AUTO_CAPTURE_MODULE,
and AUTO_CAPTURE_ACTION.
parameter_value The search criteria for the automatic capture filter.
When parameter_name is set to AUTO_CAPTURE_SQL_TEXT, the
search pattern depends on the allow setting:
• LIKE
The parameter uses this pattern when allow=>true.
• NOT LIKE
The parameter uses this pattern when allow=>false.
For all other non-null parameter_name values, the search pattern
depends on the allow setting:
• =
The parameter uses this pattern when allow=>true.
• <>
The parameter uses this pattern when allow=>false.
A null value removes the filter for parameter_name entirely.
allow Whether to include (true) or exclude (false) matching SQL
statements and plans. If null, then the procedure ignores the specified
parameter.
You can configure multiple parameters of different types. Also, you can specify
multiple values for the same parameter in separate statements, which the database
combines. The settings are additive: one parameter setting does not override a
previous setting. For example, the following filter captures SQL in the parsing schema
SYS or SYSTEM:
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_PARSING_SCHEMA_NAME','SYS',true);
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_PARSING_SCHEMA_NAME','SYSTEM',true);
However, you cannot configure multiple values for the same parameter in the same
procedure. For example, you cannot specify multiple SQL text strings for
AUTO_CAPTURE_SQL_TEXT.
29-8
Chapter 29
Configuring SQL Plan Management
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_PARSING_SCHEMA_NAME',null,true);
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_SQL_TEXT',null,true);
3. Include only statements parsed in the sh schema for consideration for automatic
capture:
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_PARSING_SCHEMA_NAME','sh',true);
4. Exclude any statement that contains the text TEST_ONLY from consideration for
automatic capture:
EXEC DBMS_SPM.CONFIGURE('AUTO_CAPTURE_SQL_TEXT','%TEST_ONLY%',false);
PARAMETER_NAME PARAMETER_VALUE
-------------------------------- --------------------------------
AUTO_CAPTURE_PARSING_SCHEMA_NAME parsing_schema IN (SH)
AUTO_CAPTURE_MODULE
AUTO_CAPTURE_ACTION
AUTO_CAPTURE_SQL_TEXT (sql_text NOT LIKE %TEST_ONLY%)
29-9
Chapter 29
Configuring SQL Plan Management
See Also:
Typically, you might want to disable one or two plan baselines, but not all of them. A
possible use case might be testing the benefits of SQL plan management.
If the parameters are set as you intend, then skip the remaining steps.
2. To ignore all existing plan baselines enter the following statement:
See Also:
Oracle Database Reference to learn about the SQL plan baseline
initialization parameters
29-10
Chapter 29
Configuring SQL Plan Management
Note:
The SPM Evolve Advisor task can accept more than one plan for a single
statement.
SPM Evolve Advisor locates unaccepted plans that the optimizer added to the SMB
during earlier hard parses. In some cases, more optimal plans may reside in other
locations. By using the DBMS_SPM.SET_EVOLVE_TASK_PARAMETER procedure, you can
configure the automatic task to search the shared SQL area, AWR repository, or SQL
tuning sets for plans that are not yet in the SMB plan history. If the advisor finds
eligible plans in alternative locations, then it includes them along with the other
unaccepted plans. By default, the automatic task does not search alternative locations.
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DBMS_SPM.SET_EVOLVE_TASK_PARAMETER procedure
29-11
Chapter 29
Configuring SQL Plan Management
See Also:
"Enabling and Disabling the Automatic SQL Tuning Task" to learn how to
enable and disable Automatic SPM Evolve Advisor
Because the SYS_AUTO_SPM_EVOLVE_TASK task is owned by SYS, only SYS can set task
parameters.
The following table describes some of the procedure parameters.
Parameter Description
alternate_plan_source Determines which sources to search for additional plans:
CURSOR_CACHE, AUTOMATIC_WORKLOAD_REPOSITORY, or
SQL_TUNING_SETS. Combine multiple values with the plus
sign (+). The default is CURSOR_CACHE
+AUTOMATIC_WORKLOAD_REPOSITORY.
alternate_plan_baseline Determines which alternative plans should be loaded.
EXISTING, which is the default, loads plans with for
statements with existing baselines. NEW loads plans for
statements without a baseline, in which case a new
baseline is created. Combine multiple values with the plus
sign (+).
alternate_plan_limit Specifies the maximum number of plans to load in total
(that is, not for each SQL statement). The default is 10.
accept_plans Specifies whether to accept recommended plans
automatically. When ACCEPT_PLANS is true (default),
SQL plan management automatically accepts all plans
recommended by the task. When set to false, the task
verifies the plans and generates a report if its findings, but
does not evolve the plans.
29-12
Chapter 29
Configuring SQL Plan Management
• You want the evolve task to look for up to a maximum of 500 plans in the shared
SQL area and AWR repository
PARAMETER_NAME VALUE
------------------------- ------------------------------------------
ALTERNATE_PLAN_LIMIT 0
ALTERNATE_PLAN_SOURCE CURSOR_CACHE+AUTOMATIC_WORKLOAD_REPOSITORY
ALTERNATE_PLAN_BASELINE EXISTING
ACCEPT_PLANS true
TIME_LIMIT 3600
BEGIN
DBMS_SPM.SET_EVOLVE_TASK_PARAMETER(
task_name => 'SYS_AUTO_SPM_EVOLVE_TASK'
, parameter => parameter_name
, value => value
);
END;
/
BEGIN
DBMS_SPM.SET_EVOLVE_TASK_PARAMETER(
task_name => 'SYS_AUTO_SPM_EVOLVE_TASK'
, parameter => 'TIME_LIMIT'
, value => '1200'
);
DBMS_SPM.SET_EVOLVE_TASK_PARAMETER(
task_name => 'SYS_AUTO_SPM_EVOLVE_TASK'
29-13
Chapter 29
Displaying Plans in a SQL Plan Baseline
4. Optionally, confirm your changes by querying the current parameter settings for
SYS_AUTO_SPM_EVOLVE_TASK.
For example, execute the following query:
PARAMETER_NAME VALUE
------------------------- ------------------------------------------
ALTERNATE_PLAN_LIMIT 500
ALTERNATE_PLAN_SOURCE CURSOR_CACHE+AUTOMATIC_WORKLOAD_REPOSITORY
ALTERNATE_PLAN_BASELINE EXISTING
ACCEPT_PLANS true
TIME_LIMIT 1200
See Also:
29-14
Chapter 29
Displaying Plans in a SQL Plan Baseline
This section explains how to show plans in a baseline from the command line. If you
use Cloud Control, then display plan baselines from the SQL Plan Baseline subpage
shown in Figure 29-1.
SELECT PLAN_TABLE_OUTPUT
FROM V$SQL s, DBA_SQL_PLAN_BASELINES b,
TABLE(
DBMS_XPLAN.DISPLAY_SQL_PLAN_BASELINE(b.sql_handle,b.plan_name,'basic')
) t
WHERE s.EXACT_MATCHING_SIGNATURE=b.SIGNATURE
AND b.PLAN_NAME=s.SQL_PLAN_BASELINE
AND s.SQL_ID='31d96zzzpcys9';
PLAN_TABLE_OUTPUT
------------------------------------------------------------------------
---
------------------------------------------------------------------------
---
SQL handle: SQL_513f7f8a91177b1a
SQL text: select * from hr.employees where employee_id=100
------------------------------------------------------------------------
---
------------------------------------------------------------------------
---
Plan name: SQL_PLAN_52gvzja8jfysuc0e983c6 Plan id: 3236529094
Enabled: YES Fixed: NO Accepted: YES Origin: AUTO-CAPTURE
------------------------------------------------------------------------
29-15
Chapter 29
Loading SQL Plan Baselines
---
-----------------------------------------------------
| Id | Operation | Name |
-----------------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | TABLE ACCESS BY INDEX ROWID| EMPLOYEES |
| 2 | INDEX UNIQUE SCAN | EMP_EMP_ID_PK |
-----------------------------------------------------
The results show that the plan for SQL ID 31d96zzzpcys is named
SQL_PLAN_52gvzja8jfysuc0e983c6 and was captured automatically.
See Also:
The goal of this task is to load plans from the following sources:
29-16
Chapter 29
Loading SQL Plan Baselines
• AWR
Load plans from Automatic Workload Repository (AWR) snapshots. You must
specify the beginning and ending of the snapshot range. Optionally, you can apply
a filter to load only plan that meet specified criteria. By default, the optimizer uses
the loaded plans the next time that the database executes the SQL statements.
• Shared SQL area
Load plans for statements directly from the shared SQL area, which is in the
shared pool of the SGA. By applying a filter on the module name, the schema, or
the SQL ID you identify the SQL statement or set of SQL statements to capture.
The optimizer uses the plans the next time that the database executes the SQL
statements.
Loading plans directly from the shared SQL area is useful when application SQL
has been hand-tuned using hints. Because you probably cannot change the SQL
to include the hint, populating the SQL plan baseline ensures that the application
SQL uses optimal plans.
• SQL tuning set (STS)
Capture the plans for a SQL workload into an STS, and then load the plans into
the SQL plan baselines. The optimizer uses the plans the next time that the
database executes the SQL statements. Bulk loading execution plans from an
STS is an effective way to prevent plan regressions after a database upgrade.
• Staging table
Use the DBMS_SPM package to define a staging table,
DBMS_SPM.PACK_STGTAB_BASELINE to copy the baselines into a staging table, and
Oracle Data Pump to transfer the table to another database. On the destination
database, use DBMS_SPM.UNPACK_STGTAB_BASELINE to unpack the plans from the
staging table and put the baselines into the SMB.
A use case is the introduction of new SQL statements into the database from a
new application module. A vendor can ship application software with SQL plan
baselines for the new SQL. In this way, the new SQL uses plans that are known to
give optimal performance under a standard test configuration. Alternatively, if you
develop or test an application in-house, export the correct plans from the test
database and import them into the production database.
• Stored outline
Migrate stored outlines to SQL plan baselines. After the migration, you maintain
the same plan stability that you had using stored outlines while being able to use
the more advanced features provided by SQL Plan Management, such as plan
evolution. See .
See Also:
29-17
Chapter 29
Loading SQL Plan Baselines
This section explains how to load plans using the command line. In Cloud Control, go
to the SQL Plan Baseline subpage (shown in Figure 29-1) and click Load to load plan
baselines from AWR.
This tutorial assumes the following:
• You want to load plans for the following query into the SMB:
SELECT /*LOAD_AWR*/ *
FROM sh.sales
WHERE quantity_sold > 40
ORDER BY prod_id;
SELECT *
FROM (SELECT SNAP_ID, SNAP_LEVEL,
TO_CHAR(BEGIN_INTERVAL_TIME, 'DD/MM/YY HH24:MI:SS') BEGIN
29-18
Chapter 29
Loading SQL Plan Baselines
FROM DBA_HIST_SNAPSHOT
ORDER BY SNAP_ID DESC)
WHERE ROWNUM <= 3;
2. Query sh.sales, using the LOAD_AWR tag to identify the SQL statement.
For example, use the following query:
SELECT /*LOAD_AWR*/ *
FROM sh.sales
WHERE quantity_sold > 40
ORDER BY prod_id;
EXEC DBMS_WORKLOAD_REPOSITORY.CREATE_SNAPSHOT;
4. Query the most recent 3 AWR snapshots to confirm that a new snapshot was
taken.
For example, query DBA_HIST_SNAPSHOT as follows:
SELECT *
FROM (SELECT SNAP_ID, SNAP_LEVEL,
TO_CHAR(BEGIN_INTERVAL_TIME, 'DD/MM/YY HH24:MI:SS') BEGIN
FROM DBA_HIST_SNAPSHOT
ORDER BY SNAP_ID DESC)
WHERE ROWNUM <= 3;
5. Load the plans for the most recent 2 snapshots from AWR.
For example, execute the LOAD_PLANS_FROM_AWR function in SQL*Plus to load the
plans from snapshot 212 to 213:
In the preceding example, the variable v_plan_cnt contains the number of plans
that were loaded.
29-19
Chapter 29
Loading SQL Plan Baselines
6. Query the data dictionary to ensure that the plans were loaded into the baseline
for the LOAD_AWR statement.
The following statement queries DBA_SQL_PLAN_BASELINES (sample output
included):
The output shows that the plan is accepted, which means that it is in the plan
baseline for the statement. Also, the origin is MANUAL-LOAD-FROM-AWR, which
means that the statement was loaded manually from AWR rather than
automatically captured.
See Also:
• "Fixed Plans"
• Oracle Database PL/SQL Packages and Types Reference to learn how
to use the DBMS_SPM.LOAD_PLANS_FROM_AWR function
• Oracle Database Reference to learn more about the
DBA_SQL_PLAN_BASELINES view
29-20
Chapter 29
Loading SQL Plan Baselines
This section explains how to load plans using the command line. In Cloud Control, go
to the SQL Plan Baseline subpage (shown in Figure 29-1) and click Load to load plan
baselines from the shared SQL area.
This tutorial assumes the following:
• You have executed the following query:
SELECT /*LOAD_CC*/ *
FROM sh.sales
WHERE quantity_sold > 40
ORDER BY prod_id;
The preceding output shows that the SQL ID of the statement is 27m0sdw9snw59.
2. Load the plans for the specified statements into the SQL plan baseline.
For example, execute the LOAD_PLANS_FROM_CURSOR_CACHE function in SQL*Plus to
load the plan for the statement with the SQL ID 27m0sdw9snw59:
29-21
Chapter 29
Loading SQL Plan Baselines
In the preceding example, the variable v_plan_cnt contains the number of plans
that were loaded.
3. Query the data dictionary to ensure that the plans were loaded into the baseline
for the statement.
The following statement queries DBA_SQL_PLAN_BASELINES (sample output
included):
The output shows that the plan is accepted, which means that it is in the plan
baseline for the statement. Also, the origin is MANUAL-LOAD-FROM-CC, which means
that the statement was loaded manually from the shared SQL area rather than
automatically captured.
See Also:
• "Fixed Plans"
• Oracle Database PL/SQL Packages and Types Reference to learn how
to use the DBMS_SPM.LOAD_PLANS_FROM_CURSOR_CACHE function
• Oracle Database Reference to learn more about the
DBA_SQL_PLAN_BASELINES view
29-22
Chapter 29
Loading SQL Plan Baselines
Function Description
Parameter
sqlset_name Name of the STS from which the plans are loaded into SQL plan baselines.
basic_filter A filter applied to the STS to select only qualifying plans to be loaded. The
filter can take the form of any WHERE clause predicate that can be specified
against the view DBA_SQLSET_STATEMENTS. An example is basic_filter
=> 'sql_text like ''SELECT /*LOAD_STS*/%'''.
fixed Default NO means the loaded plans are used as nonfixed plans. YES means
the loaded plans are fixed plans. "Plan Selection" explains that the
optimizer chooses a fixed plan in the plan baseline over a nonfixed plan.
This section explains how to load plans from the command line. In Cloud Control, go to
the SQL Plan Baseline subpage (shown in Figure 29-1) and click Load to load plan
baselines from SQL tuning sets.
This tutorial assumes the following:
• You want the loaded plans to be nonfixed.
• You have executed the following query:
SELECT /*LOAD_STS*/ *
FROM sh.sales
WHERE quantity_sold > 40
ORDER BY prod_id;
• You have loaded the plan from the shared SQL area into the SQL tuning set
named SPM_STS, which is owned by user SPM.
SELECT SQL_TEXT
FROM DBA_SQLSET_STATEMENTS
WHERE SQLSET_NAME = 'SPM_STS';
SQL_TEXT
--------------------
SELECT /*LOAD_STS*/
*
FROM sh.sales
WHERE quantity_sold
> 40
ORDER BY prod_id
29-23
Chapter 29
Loading SQL Plan Baselines
The output shows that the plan for the select /*LOAD_STS*/ statement is in the
STS.
2. Load the plan from the STS into the SQL plan baseline.
For example, in SQL*Plus execute the function as follows:
The basic_filter parameter specifies a WHERE clause that loads only the plans for
the queries of interest. The variable v_plan_cnt stores the number of plans loaded
from the STS.
3. Query the data dictionary to ensure that the plan was loaded into the baseline for
the statement.
The following statement queries the DBA_SQL_PLAN_BASELINES view (sample output
included).
The output shows that the plan is accepted, which means that it is in the plan
baseline. Also, the origin is MANUAL-LOAD-FROM-STS, which means that the plan
was loaded manually from a SQL tuning set rather than automatically captured.
4. Optionally, drop the STS.
For example, execute DBMS_SQLTUNE.DROP_SQLSET to drop the SPM_STS tuning set
as follows:
29-24
Chapter 29
Loading SQL Plan Baselines
See Also:
Pack 2
1 Unpack 6
1
SQL SQL
Management Management
Base Base
SQL Plan SQL Plan
Baselines Baselines
29-25
Chapter 29
Loading SQL Plan Baselines
Export plans with the PACK_STGTAB_BASELINE function of the DBMS_SPM package, and
then import them with UNPACK_STGTAB_BASELINE. The following table describes some
function parameters.
BEGIN
DBMS_SPM.CREATE_STGTAB_BASELINE (
table_name => 'stage1');
END;
/
2. On the source database, pack the SQL plan baselines you want to export from the
SQL management base into the staging table.
The following example packs enabled plan baselines created by user spm into
staging table stage1. Select SQL plan baselines using the plan name (plan_name),
SQL handle (sql_handle), or any other plan criteria. The table_name parameter is
mandatory.
DECLARE
v_plan_cnt NUMBER;
BEGIN
v_plan_cnt := DBMS_SPM.PACK_STGTAB_BASELINE (
table_name => 'stage1'
, enabled => 'yes'
, creator => 'spm'
);
29-26
Chapter 29
Evolving SQL Plan Baselines Manually
END;
/
3. Export the staging table stage1 into a dump file using Oracle Data Pump Export.
4. Transfer the dump file to the host of the destination database.
5. On the destination database, import the staging table stage1 from the dump file
using the Oracle Data Pump Import utility.
6. On the destination database, unpack the SQL plan baselines from the staging
table into the SQL management base.
The following example unpacks all fixed plan baselines stored in the staging table
stage1:
DECLARE
v_plan_cnt NUMBER;
BEGIN
v_plan_cnt := DBMS_SPM.UNPACK_STGTAB_BASELINE (
table_name => 'stage1'
, fixed => 'yes'
);
END;
/
See Also:
29-27
Chapter 29
Evolving SQL Plan Baselines Manually
See Also:
"Managing the SPM Evolve Advisor Task"
Table 29-9 DBMS_SPM Functions and Procedures for Managing Plan Evolution
Tasks
Oracle recommends that you configure SPM Evolve Advisor to run automatically. You
can also evolve SQL plan baselines manually. The following graphic shows the basic
workflow for managing SQL plan management tasks.
29-28
Chapter 29
Evolving SQL Plan Baselines Manually
DBMS_SPM.CREATE_EVOLVE_TASK
DBMS_SPM.SET_EVOLVE_TASK_PARAMETER
DBMS_SPM.EXECUTE_EVOLVE_TASK
DBMS_SPM.IMPLEMENT_EVOLVE_TASK
DBMS_SPM.REPORT_EVOLVE_TASK
Typically, you manage SQL plan evolution tasks in the following sequence:
1. Create an evolve task
2. Optionally, set evolve task parameters
3. Execute the evolve task
4. Implement the recommendations in the task
5. Report on the task outcome
See Also:
29-29
Chapter 29
Evolving SQL Plan Baselines Manually
This section explains how to evolve plan baselines from the command line. In Cloud
Control, from the SQL Plan Baseline subpage, select a plan, and then click Evolve.
This tutorial assumes the following:
• You do not have the automatic evolve task enabled.
• You want to create a SQL plan baseline for the following query:
• You want to create two indexes to improve the query performance, and then
evolve the plan that uses these indexes if it performs better than the plan currently
in the plan baseline.
c. Connect to the database as user sh, and then set SQL*Plus display
parameters:
CONNECT sh
-- enter password
29-30
Chapter 29
Evolving SQL Plan Baselines Manually
2. Execute the SELECT statements so that SQL plan management captures them:
a. Execute the SELECT /* q1_group_by */ statement for the first time.
Because the database only captures plans for repeatable statements, the plan
baseline for this statement is empty.
b. Query the data dictionary to confirm that no plans exist in the plan baseline.
For example, execute the following query (sample output included):
no rows selected
29-31
Chapter 29
Evolving SQL Plan Baselines Manually
The output shows that the plan is accepted, which means that it is in the plan
baseline for the statement. Also, the origin is AUTO-CAPTURE, which means that the
statement was automatically captured and not manually loaded.
4. Explain the plan for the statement and verify that the optimizer is using this plan.
For example, explain the plan as follows, and then display it:
------------------------------------------
| Id | Operation | Name |
------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | HASH GROUP BY | |
| 2 | HASH JOIN | |
| 3 | TABLE ACCESS FULL | PRODUCTS |
| 4 | PARTITION RANGE ALL| |
| 5 | TABLE ACCESS FULL | SALES |
------------------------------------------
Note
-----
- SQL plan baseline "SQL_PLAN_0gwbcfvzskcu242949306" used for this
statement
The note indicates that the optimizer is using the plan shown with the plan name
listed in the previous step.
5. Create two indexes to improve the performance of the SELECT /* q1_group_by */
statement.
For example, use the following statements:
29-32
Chapter 29
Evolving SQL Plan Baselines Manually
Because automatic capture is enabled, the plan baseline is populated with the new
plan for this statement.
7. Query the data dictionary to ensure that the plan was loaded into the SQL plan
baseline for the statement.
The following statement queries DBA_SQL_PLAN_BASELINES (sample output
included).
The output shows that the new plan is unaccepted, which means that it is in the
statement history but not the SQL plan baseline.
8. Explain the plan for the statement and verify that the optimizer is using the original
unindexed plan.
For example, explain the plan as follows, and then display it:
29-33
Chapter 29
Evolving SQL Plan Baselines Manually
GROUP BY prod_name;
SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY(null, null, 'basic +note'));
------------------------------------------
| Id | Operation | Name |
------------------------------------------
| 0 | SELECT STATEMENT | |
| 1 | HASH GROUP BY | |
| 2 | HASH JOIN | |
| 3 | TABLE ACCESS FULL | PRODUCTS |
| 4 | PARTITION RANGE ALL| |
| 5 | TABLE ACCESS FULL | SALES |
------------------------------------------
Note
-----
- SQL plan baseline "SQL_PLAN_0gwbcfvzskcu242949306" used for this
statement
The note indicates that the optimizer is using the plan shown with the plan name
listed in Step 3.
9. Connect as an administrator, and then create an evolve task that considers all
SQL statements with unaccepted plans.
For example, execute the DBMS_SPM.CREATE_EVOLVE_TASK function and then obtain
the name of the task:
CONNECT / AS SYSDBA
VARIABLE cnt NUMBER
VARIABLE tk_name VARCHAR2(50)
VARIABLE exe_name VARCHAR2(50)
VARIABLE evol_out CLOB
:EVOL_OUT
------------------------------------------------------------------------
--
TASK_11
Now that the task has been created and has a unique name, execute the task.
10. Execute the task.
29-34
Chapter 29
Evolving SQL Plan Baselines Manually
:EXE_NAME
------------------------------------------------------------------------
---
EXEC_1
Task Information:
---------------------------------------------
Task Name : TASK_11
Task Owner : SYS
Execution Name : EXEC_1
Execution Type : SPM EVOLVE
Scope : COMPREHENSIVE
Status : COMPLETED
Started : 01/09/2012 12:21:27
Finished : 01/09/2012 12:21:29
Last Updated : 01/09/2012 12:21:29
Global Time Limit : 2147483646
Per-Plan Time Limit : UNUSED
Number of Errors : 0
------------------------------------------------------------------------
---
SUMMARY SECTION
------------------------------------------------------------------------
---
Number of plans processed : 1
Number of findings : 1
Number of recommendations : 1
Number of errors : 0
------------------------------------------------------------------------
---
DETAILS SECTION
------------------------------------------------------------------------
---
Object ID : 2
29-35
Chapter 29
Evolving SQL Plan Baselines Manually
Execution Statistics:
-----------------------------
Base Plan Test Plan
----------------------------
------------------------
Elapsed Time (s): .044336 .012649
CPU Time (s): .044003 .012445
Buffer Gets: 360 99
Optimizer Cost: 924 891
Disk Reads: 341 82
Direct Writes: 0 0
Rows Processed: 4 2
Executions: 5 9
FINDINGS SECTION
------------------------------------------------------------------------
---
Findings (1):
-----------------------------
1. The plan was verified in 2.18 seconds. It passed the benefit
criterion
because its verified performance was 2.01 times better than that of
the
baseline plan.
Recommendation:
-----------------------------
Consider accepting the plan. Execute
dbms_spm.accept_sql_plan_baseline(task_name => 'TASK_11', object_id =>
2,
task_owner => 'SYS');
Baseline Plan
-----------------------------
Plan Id : 1
Plan Hash Value : 1117033222
29-36
Chapter 29
Evolving SQL Plan Baselines Manually
------------------------------------------------------------------------
---
| Id| Operation | Name | Rows | Bytes |Cost |
Time |
------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | | 21 | 861 | 924 |
00:00:12|
| 1 | HASH GROUP BY | | 21 | 861 | 924 |
00:00:12|
| *2| HASH JOIN | |267996|10987836 | 742 |
00:00:09|
| *3| TABLE ACCESS FULL | PRODUCTS | 21 | 714 | 2 |
00:00:01|
| 4 | PARTITION RANGE ALL | |918843| 6431901 | 662 |
00:00:08|
| 5 | TABLE ACCESS FULL | SALES |918843| 6431901 | 662 |
00:00:08|
------------------------------------------------------------------------
---
Test Plan
-----------------------------
Plan Id : 2
Plan Hash Value : 20315500
------------------------------------------------------------------------
---
|Id| Operation | Name | Rows | Bytes | Cost|
Time |
------------------------------------------------------------------------
---
| 0|SELECT STATEMENT | | 21| 861|891|
00:00:11|
| 1| SORT GROUP BY NOSORT| | 21| 861|891|
00:00:11|
| 2| NESTED LOOPS | |267996|10987836|891|
00:00:11|
|*3| INDEX RANGE SCAN |IND_PROD_CAT_NAME | 21| 714| 1|
00:00:01|
|*4| INDEX RANGE SCAN |IND_SALES_PROD_QTY| 12762| 89334| 42|
00:00:01|
------------------------------------------------------------------------
---
29-37
Chapter 29
Evolving SQL Plan Baselines Manually
This report indicates that the new execution plan, which uses the two new indexes,
performs better than the original plan.
12. Implement the recommendations of the evolve task.
13. Query the data dictionary to ensure that the new plan is accepted.
29-38
Chapter 29
Dropping SQL Plan Baselines
See Also:
Function Description
Parameter
sql_handle SQL statement identifier.
plan_name Name of a specific plan. Default NULL drops all plans associated with the
SQL statement identified by sql_handle.
This section explains how to drop baselines from the command line. In Cloud Control,
from the SQL Plan Baseline subpage, select a plan, and then click Drop.
This tutorial assumes that you want to drop all plans for the following SQL statement,
effectively dropping the SQL plan baseline:
29-39
Chapter 29
Managing the SQL Management Base
DECLARE
v_dropped_plans number;
BEGIN
v_dropped_plans := DBMS_SPM.DROP_SQL_PLAN_BASELINE (
sql_handle => 'SQL_b6b0d1c71cd1807b'
);
DBMS_OUTPUT.PUT_LINE('dropped ' || v_dropped_plans || ' plans');
END;
/
no rows selected
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn about the
DROP_SQL_PLAN_BASELINE function
29-40
Chapter 29
Managing the SQL Management Base
Parameter Description
SPACE_BUDGET_PERCENT Maximum percent of SYSAUX space that the SQL
management base can use. The default is 10. The
allowable range for this limit is between 1% and 50%.
PLAN_RETENTION_WEEKS Number of weeks to retain unused plans before they are
purged. The default is 53.
AUTO_CAPTURE_PARSING_SCHEM A list of the form (% LIKE a OR % LIKE b ...) AND (%
A_NAME NOT LIKE c AND % NOT LIKE d ...), which is the
internal representation of the parsing schema name filter. If
no parsing schema filters exist, then one side of the outer
conjunction will be absent.
AUTO_CAPTURE_MODULE A list of the form (% LIKE a OR % LIKE b ...) AND (%
NOT LIKE c AND % NOT LIKE d ...), which is the
internal representation of the module filter. If no module
filters exist, then one side of the outer conjunction will be
absent.
AUTO_CAPTURE_ACTION A list of the form (% LIKE a OR % LIKE b ...) AND (%
NOT LIKE c AND % NOT LIKE d ...), which is the
internal representation of the action filter. If no action filters
exist, then one side of the outer conjunction will be absent.
AUTO_CAPTURE_SQL_TEXT A list of the form (% LIKE a OR % LIKE b ...) AND (%
NOT LIKE c AND % NOT LIKE d ...), which is the
internal representation of the SQL text filter. If no SQL text
filters exist, then one side of the outer conjunction will be
absent.
29-41
Chapter 29
Managing the SQL Management Base
See Also:
Assumptions
This tutorial assumes the following:
• The current SMB space limit is the default of 10%.
• You want to change the percentage limit to 30%
EXECUTE DBMS_SPM.CONFIGURE('space_budget_percent',30);
29-42
Chapter 29
Managing the SQL Management Base
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the DBMS_SPM.CONFIGURE procedure
PARAMETER_NAME PARAMETER_VALUE
------------------------------ ---------------
PLAN_RETENTION_WEEKS 53
29-43
Chapter 29
Managing the SQL Management Base
For example, execute the CONFIGURE procedure to change the period to 105
weeks:
EXECUTE DBMS_SPM.CONFIGURE('plan_retention_weeks',105);
PARAMETER_NAME PARAMETER_VALUE
------------------------------ ---------------
PLAN_RETENTION_WEEKS 105
See Also:
Oracle Database PL/SQL Packages and Types Reference to learn more
about the CONFIGURE procedure
29-44
30
Migrating Stored Outlines to SQL Plan
Baselines
Stored outline migration is the user-initiated process of converting stored outlines to
SQL plan baselines. A SQL plan baseline is a set of plans proven to provide optimal
performance.
This chapter explains the concepts and tasks relating to stored outline migration.
Note:
Starting in Oracle Database 12c, stored outlines are deprecated. See
"Migrating Stored Outlines to SQL Plan Baselines " for an alternative.
30-1
Chapter 30
About Stored Outline Migration
30-2
Chapter 30
About Stored Outline Migration
30-3
Chapter 30
About Stored Outline Migration
The equivalent of a category for an outline is a module for a SQL plan baseline.
Table 30-1 explains how outline categories map to modules.
30-4
Chapter 30
About Stored Outline Migration
When migrating stored outlines to SQL plan baselines, Oracle Database maps every
outline category to a SQL plan baseline module with the same name. As shown in the
following diagram, the outline category OLTP is mapped to the baseline module OLTP.
After migration, DEFAULT is a super-category that contains all SQL plan baselines.
Category DEFAULT
SELECT...
Category DW Module DW
SELECT...
30-5
Chapter 30
About Stored Outline Migration
You can control stored outline and plan baseline behavior with initialization and
session parameters. Table 30-3 describes the relevant parameters. See Table 30-5
and Table 30-6 for an explanation of how these parameter settings interact.
You can use database views to access information relating to stored outline migration.
Table 30-4 describes the following main views.
View Description
DBA_OUTLINES Describes all stored outlines in the database.
The MIGRATED column is important for outline migration and
shows one of the following values: NOT-MIGRATED and
MIGRATED. When MIGRATED, the stored outline has been
migrated to a plan baseline and is not usable.
30-6
Chapter 30
Preparing for Stored Outline Migration
View Description
DBA_SQL_PLAN_BASELINES Displays information about the SQL plan baselines currently
created for specific SQL statements.
The ORIGIN column indicates how the plan baseline was
created. The value STORED-OUTLINE indicates the baseline
was created by migrating an outline.
See Also:
30-7
Chapter 30
Preparing for Stored Outline Migration
3. Determine which stored outlines meet the following prerequisites for migration
eligibility:
• The statement must not be a run-time INSERT AS SELECT statement.
• The statement must not reference a remote object.
• This statement must not be a private stored outline.
4. Decide whether to migrate all outlines, specified stored outlines, or outlines
belonging to a specified outline category.
If you do not decide to migrate all outlines, then identify the outlines or categories
that you intend to migrate.
5. Decide whether the stored outlines migrated to SQL plan baselines use fixed
plans or nonfixed plans:
• Fixed plans
A fixed plan is frozen. If a fixed plan is reproducible using the hints stored in
plan baseline, then the optimizer always chooses the lowest-cost fixed plan
baseline over plan baselines that are not fixed. Essentially, a fixed plan
baseline acts as a stored outline with valid hints.
A fixed plan is reproducible when the database can parse the statement based
on the hints stored in the plan baseline and create a plan with the same plan
hash value as the one in the plan baseline. If one of more of the hints become
invalid, then the database may not be able to create a plan with the same plan
hash value. In this case, the plan is nonreproducible.
If a fixed plan cannot be reproduced when parsed using its hints, then the
optimizer chooses a different plan, which can be either of the following:
– Another plan for the SQL plan baseline
– The current cost-based plan created by the optimizer
In some cases, a performance regression occurs because of the different plan,
requiring SQL tuning.
• Nonfixed plans
If a plan baseline does not contain fixed plans, then SQL Plan Management
considers the plans equally when picking a plan for a SQL statement.
6. Before beginning the actual migration, ensure that the Oracle database meets the
following prerequisites:
• The database must be Enterprise Edition.
• The database must be open and not in a suspended state.
• The database must not be in restricted access (DBA), read-only, or migrate
mode.
30-8
Chapter 30
Migrating Outlines to Utilize SQL Plan Management Features
See Also:
Assumptions
This tutorial assumes the following:
• You migrate all outlines.
To migrate specific outlines, use the DBMS_SPM.MIGRATE_STORED_OUTLINE function.
• You want the module names of the baselines to be identical to the category names
of the migrated outlines.
• You do not want the SQL plans to be fixed.
By default, generated plans are not fixed and SQL Plan Management considers all
plans equally when picking a plan for a SQL statement. This situation permits the
advanced feature of plan evolution to capture new plans for a SQL statement,
verify their performance, and accept these new plans into the plan baseline.
DECLARE
my_report CLOB;
BEGIN
my_outlines := DBMS_SPM.MIGRATE_STORED_OUTLINE(
attribute_name => 'all' );
END;
/
30-9
Chapter 30
Migrating Outlines to Preserve Stored Outline Behavior
See Also:
Assumptions
This tutorial assumes the following:
• You want to migrate only the stored outlines in the category named firstrow.
• You want the module names of the baselines to be identical to the category names
of the migrated outlines.
DECLARE
my_report CLOB;
BEGIN
my_outlines := DBMS_SPM.MIGRATE_STORED_OUTLINE(
attribute_name => 'category',
attribute_value => 'firstrow',
fixed => 'YES' );
END;
/
After migration, the SQL plan baselines is in module firstrow and category
DEFAULT.
30-10
Chapter 30
Performing Follow-Up Tasks After Stored Outline Migration
See Also:
30-11
Chapter 30
Performing Follow-Up Tasks After Stored Outline Migration
30-12
Chapter 30
Performing Follow-Up Tasks After Stored Outline Migration
Table 30-6 (Cont.) Use of Stored Outlines and SQL Plan Baselines
Assumptions
This tutorial assumes the following:
• You have completed the basic steps in the stored outline migration.
• Some stored outlines may have been created before Oracle Database 10g.
Hints in releases before Oracle Database 10g use a local hint format. After
migration, hints stored in a plan baseline use the global hints format introduced in
Oracle Database 10g.
30-13
Chapter 30
Performing Follow-Up Tasks After Stored Outline Migration
DECLARE
v_cnt PLS_INTEGER;
BEGIN
v_cnt := DBMS_SPM.ALTER_SQL_PLAN_BASELINE(
sql_handle=>'SYS_SQL_f44779f7089c8fab',
attribute_name=>'FIXED',
attribute_value=>'NO');
DBMS_OUTPUT.PUT_LINE('Plans altered: ' || v_cnt);
END;
/
4. Drop all stored outlines that have been migrated to SQL plan baselines.
30-14
Chapter 30
Performing Follow-Up Tasks After Stored Outline Migration
For example, the following statements drops all stored outlines with status
MIGRATED in DBA_OUTLINES:
DECLARE
v_cnt PLS_INTEGER;
BEGIN
v_cnt := DBMS_SPM.DROP_MIGRATED_STORED_OUTLINE();
DBMS_OUTPUT.PUT_LINE('Migrated stored outlines dropped: ' || v_cnt);
END;
/
ALTER SYSTEM
SET CREATE_STORED_OUTLINE = false SCOPE = BOTH;
ALTER SYSTEM
SET OPTIMIZER_CAPTURE_SQL_PLAN_BASELINES = true SCOPE = BOTH;
ALTER SYSTEM
SET OPTIMIZER_USE_SQL_PLAN_BASELINES = true SCOPE = BOTH;
ALTER SESSION
SET USE_STORED_OUTLINES = allrows SCOPE = BOTH;
See Also:
30-15
A
Guidelines for Indexes and Table Clusters
This appendix provides an overview of data access methods using indexes and
clusters that can enhance or degrade performance.
This appendix contains the following sections:
• Guidelines for Tuning Index Performance
Indexes are not a performance cure-all. You must consider carefully when and
how to use them.
• Guidelines for Using Function-Based Indexes for Performance
A function-based index includes columns that are either transformed by a function,
such as the UPPER function, or included in an expression, such as col1 + col2.
• Guidelines for Using Partitioned Indexes for Performance
Similar to partitioned tables, partitioned indexes improve manageability,
availability, performance, and scalability. They can be partitioned independently
(global indexes) or automatically linked to the table partitioning method (local
indexes).
• Guidelines for Using Index-Organized Tables for Performance
An index-organized table differs from an ordinary table because the data for the
table is held in its associated index.
• Guidelines for Using Bitmap Indexes for Performance
Bitmap indexes can substantially improve performance of specific queries.
• Guidelines for Using Bitmap Join Indexes for Performance
In addition to a bitmap index on a single table, you can create a bitmap join index,
which is a bitmap index for the join of two or more tables.
• Guidelines for Using Domain Indexes for Performance
Domain indexes are built using the indexing logic supplied by a user-defined
indextype. An indextype is an object that specifies the routines that manage a
domain (application-specific) index.
• Guidelines for Using Table Clusters
A table cluster is a group of one or more tables that are physically stored together
because they share common columns and usually appear together in SQL
statements.
• Guidelines for Using Hash Clusters for Performance
Hash clusters group table data by applying a hash function to each row's cluster
key value.
A-1
Appendix A
Guidelines for Tuning Index Performance
A-2
Appendix A
Guidelines for Tuning Index Performance
use these indexes when the application is run. If you create new indexes to tune a
statement that is currently parsed, then Oracle Database invalidates the statement.
When the statement is next parsed, the optimizer automatically chooses a new
execution plan that could potentially use the new index. If you create new indexes on a
remote database to tune a distributed statement, then the optimizer considers these
indexes when the statement is next parsed.
Creating an index to tune one statement can affect the optimizer's choice of execution
plans for other statements. For example, if you create an index to be used by one
statement, then the optimizer can choose to use that index for other statements in the
application as well. For this reason, reexamine the application's performance and
execution plans, and rerun the SQL trace facility after you have tuned those
statements that you initially identified for tuning.
See Also:
Note:
Oracle Database automatically creates indexes, or uses existing
indexes, on the keys and expressions of unique and primary keys that
you define with integrity constraints.
Indexing low selectivity columns can be helpful when the data distribution is
skewed so that one or two values occur much less often than other values.
• Do not use standard B-tree indexes on keys or expressions with few distinct
values. Such keys or expressions are usually unselective and therefore do not
optimize performance unless the frequently selected key values appear less
frequently than the other key values. You can use bitmap indexes effectively in
such cases, unless the index is modified frequently, as in a high concurrency
OLTP application.
A-3
Appendix A
Guidelines for Tuning Index Performance
• Do not index frequently modified columns. UPDATE statements that modify indexed
columns and INSERT and DELETE statements that modify indexed tables take
longer than if there were no index. Such SQL statements must modify data in
indexes and data in tables. They also create additional undo and redo.
• Do not index keys that appear only in WHERE clauses with functions or operators. A
WHERE clause that uses a function, other than MIN or MAX, or an operator with an
indexed key does not make available the access path that uses the index except
with function-based indexes.
• Consider indexing foreign keys of referential integrity constraints in cases in which
many concurrent INSERT, UPDATE, and DELETE statements access the parent and
child tables. Such an index allows UPDATEs and DELETEs on the parent table
without share locking the child table.
• When choosing to index a key, consider whether the performance gain for queries
is worth the performance loss for INSERT, UPDATE, and DELETE statements and the
use of the space required to store the index. You might want to experiment by
comparing the processing times of the SQL statements with and without indexes.
You can measure processing time with the SQL trace facility.
See Also:
Oracle Database Concepts for more information about the effects of foreign
keys on locking
Note:
This is no longer the case with index skip scans.
A-4
Appendix A
Guidelines for Tuning Index Performance
A leading portion of an index is a set of one or more columns that were specified first
and consecutively in the list of columns in the CREATE INDEX statement that created the
index. Consider this CREATE INDEX statement:
• x, xy, and xyz combinations of columns are leading portions of the index
• yz, y, and z combinations of columns are not leading portions of the index
This section contains the following topics:
See Also:
"Index Skip Scans"
A-5
Appendix A
Guidelines for Tuning Index Performance
See Also:
"Influencing the Optimizer " for more information about the NO_INDEX, FULL,
INDEX, and INDEX_COMBINE and hints
A-6
Appendix A
Guidelines for Tuning Index Performance
then the index could actually be larger than the table. In this case, it is faster to use the
base table rather than the index to re-create the index.
To reorganize or compact an existing index or to change its storage characteristics,
use the ALTER INDEX . . . REBUILD statement. The REBUILD statement uses the
existing index as the basis for the new one. All index storage statements are
supported, such as STORAGE (for extent allocation), TABLESPACE (to move the index to a
new tablespace), and INITRANS (to change the initial number of entries).
Usually, ALTER INDEX . . . REBUILD is faster than dropping and re-creating an index,
because this statement uses the fast full scan feature. It reads all the index blocks
using multiblock I/O, then discards the branch blocks. A further advantage of this
approach is that the old index is still available for queries while the rebuild is in
progress.
See Also:
Oracle Database SQL Language Reference for more information about the
CREATE INDEX and ALTER INDEX statements and restrictions on rebuilding
indexes
This option enables you to combine leaf levels of an index to free blocks for reuse.
You can also rebuild the index online.
See Also:
Oracle Database SQL Language Reference and Oracle Database
Administrator’s Guide for more information about the syntax for this
statement
The advantage of this approach is that the index remains available and valid when the
constraint is disabled. Therefore, enabling a disabled UNIQUE or PRIMARY KEY
constraint does not require rebuilding the unique index associated with the constraint.
This technique can yield significant time savings on enable operations for large tables.
Using a nonunique index to enforce uniqueness also enables you to eliminate
redundant indexes. You do not need a unique index on a primary key column if that
column is included as the prefix of a composite index. You can use the existing index
to enable and enforce the constraint. You also save significant space by not
A-7
Appendix A
Guidelines for Tuning Index Performance
duplicating the index. However, if the existing index is partitioned, then the partitioning
key of the index must also be a subset of the UNIQUE key; otherwise, Oracle Database
creates an additional unique index to enforce the constraint.
Note:
By default, constraints are created in the ENABLED state.
A-8
Appendix A
Guidelines for Using Function-Based Indexes for Performance
At this point, users can start performing INSERT, UPDATE, DELETE, and SELECT
operations on table t.
See Also:
Oracle Database Concepts for an overview of integrity constraints
SELECT *
FROM employees
WHERE UPPER(last_name) = 'MARKSON';
A-9
Appendix A
Guidelines for Using Partitioned Indexes for Performance
See Also:
See Also:
Oracle Database Concepts and Oracle Database Administrator’s Guide for
more information about global indexes tables
A-10
Appendix A
Guidelines for Using Index-Organized Tables for Performance
See Also:
A-11
Appendix A
Guidelines for Using Bitmap Join Indexes for Performance
• The bitmap indexes used in the queries have been created on some or all of these
low-cardinality or medium-cardinality columns.
• The tables in the queries contain many rows.
You can use multiple bitmap indexes to evaluate the conditions on a single table.
Bitmap indexes are thus highly advantageous for complex ad hoc queries that contain
lengthy WHERE clauses. Bitmap indexes can also provide optimal performance for
aggregate queries and for optimizing joins in star schemas.
See Also:
See Also:
Oracle Database Data Warehousing Guide for examples and restrictions of
bitmap join indexes
A-12
Appendix A
Guidelines for Using Table Clusters
The cartridge determines the parameters you can specify in creating and maintaining
the domain index. Similarly, the performance and storage characteristics of the domain
index are presented in the specific cartridge documentation.
Refer to the appropriate cartridge documentation for information such as the following:
• Which data types can be indexed?
• Which indextypes are provided?
• Which operators does the indextype support?
• How can the domain index be created and maintained?
• What is the most efficient way to use the operator in queries?
• What are the performance characteristics?
Note:
You can also create index types with the CREATE INDEXTYPE statement.
See Also:
Oracle Spatial and Graph Developer's Guide to learn more about indexing
spatial data
A-13
Appendix A
Guidelines for Using Hash Clusters for Performance
• The application joins the tables only occasionally or modifies their common column
values frequently.
Modifying a row's cluster key value takes longer than modifying the value in an
nonclustered table, because Oracle Database might need to migrate the modified
row to another block to maintain the cluster.
• The application often performs full table scans of only one of the tables.
A full table scan of a clustered table can take longer than a full table scan of an
nonclustered table. Oracle Database is likely to read more blocks because the
tables are stored together.
• The data from all tables with the same cluster key value exceeds more than one or
two data blocks.
To access a row in a clustered table, Oracle Database reads all blocks containing
rows with that value. If these rows take up multiple blocks, then accessing a single
row could require more reads than accessing the same row in a nonclustered
table.
• The number of rows for each cluster key value varies significantly.
This causes waste of space for the low cardinality key value. It causes collisions
for the high cardinality key values. Collisions degrade performance.
Consider the benefits and drawbacks of clusters for the application. For example, you
might decide that the performance gain for join statements outweighs the performance
loss for statements that modify cluster key values. You might want to experiment and
compare processing times with the tables both clustered and stored separately.
See Also:
A-14
Appendix A
Guidelines for Using Hash Clusters for Performance
• Store a table in a hash cluster if you can determine how much space is required to
hold all rows with a given cluster key value, including rows to be inserted
immediately and rows to be inserted in the future.
• Use sorted hash clusters, where rows corresponding to each value of the hash
function are sorted on a specific columns in ascending order, when the database
can improve response time on operations with this sorted clustered data.
• Do not store a table in a hash cluster in the following cases:
– The application often performs full table scans.
– You must allocate a great deal of space to the hash cluster in anticipation of
the table growing.
Full table scans must read all blocks allocated to the hash cluster, even though
some blocks might contain few rows. Storing the table alone reduces the number
of blocks read by full table scans.
• Do not store a table in a hash cluster if the application frequently modifies the
cluster key values. Modifying a row's cluster key value can take longer than
modifying the value in an nonclustered table, because Oracle Database might
need to migrate the modified row to another block to maintain the cluster.
If hashing is appropriate for the table based on the considerations in this list, then
storing a single table in a hash cluster can be useful. This is true regardless of whether
the table is joined frequently with other tables.
See Also:
A-15
Glossary
accepted plan
In the context of SQL plan management, a plan that is in a SQL plan baseline for a
SQL statement and thus available for use by the optimizer. An accepted plan contains
a set of hints, a plan hash value, and other plan-related information.
access path
The means by which the database retrieves data from a database. For example, a
query using an index and a query using a full table scan use different access paths.
adaptive optimizer
A feature of the optimizer that enables it to adapt plans based on run-time statistics.
ADDM
See Automatic Database Diagnostic Monitor (ADDM).
Glossary-1
Glossary
antijoin
A join that returns rows that fail to match the subquery on the right side. For example,
an antijoin can list departments with no employees. Antijoins use the NOT EXISTS or
NOT IN constructs.
automatic reoptimization
The ability of the optimizer to automatically change a plan on subsequent executions
of a SQL statement. Automatic reoptimization can fix any suboptimal plan chosen due
to incorrect optimizer estimates, from a suboptimal distribution method to an incorrect
choice of degree of parallelism.
Glossary-2
Glossary
AWR
See Automatic Workload Repository (AWR).
AWR snapshot
A set of data for a specific time that is used for performance comparisons. The delta
values captured by the snapshot represent the changes for each statistic over the time
period. Statistics gathered by are queried from memory. You can display the gathered
data in both reports and views.
band join
A special type of nonequijoin in which key values in one data set must fall within the
specified range (“band”) of the second data set.
base cardinality
For a table, the total number of rows in the table.
baseline
In the context of AWR, the interval between two AWR snapshots that represent the
database operating at an optimal level.
bind-aware cursor
A bind-sensitive cursor that is eligible to use different plans for different bind values.
After a cursor has been made bind-aware, the optimizer chooses plans for future
executions based on the bind value and its cardinality estimate.
bind-sensitive cursor
A cursor whose optimal plan may depend on the value of a bind variable. The
database monitors the behavior of a bind-sensitive cursor that uses different bind
values to determine whether a different plan is beneficial.
bind variable
A placeholder in a SQL statement that must be replaced with a valid value or value
address for the statement to execute successfully. By using bind variables, you can
Glossary-3
Glossary
write a SQL statement that accepts inputs or parameters at run time. The following
query uses v_empid as a bind variable:
bitmap piece
A subcomponent of a single bitmap index entry. Each indexed column value may have
one or more bitmap pieces. The database uses bitmap pieces to break up an index
entry that is large in relation to the size of a block.
B-tree index
An index organized like an upside-down tree. A B-tree index has two types of blocks:
branch blocks for searching and leaf blocks that store values. The leaf blocks contain
every indexed data value and a corresponding rowid used to locate the actual row.
The "B" stands for "balanced" because all leaf blocks automatically stay at the same
depth.
bulk load
A CREATE TABLE AS SELECT or INSERT INTO ... SELECT operation.
cardinality
The number of rows that is expected to be or is returned by an operation in an
execution plan.
Cartesian join
A join in which one or more of the tables does not have any join conditions to any
other tables in the statement. The optimizer joins every row from one data source with
every row from the other data source, creating the Cartesian product of the two sets.
child cursor
The cursor containing the plan, compilation environment, and other information for a
statement whose text is stored in a parent cursor. The parent cursor is number 0, the
first child is number 1, and so on. Child cursors reference the same SQL text as the
Glossary-4
Glossary
parent cursor, but are different. For example, two queries with the text SELECT * FROM
t use different cursors when they reference two different tables named t.
cluster scan
An access path for a table cluster. In an indexed table cluster, Oracle Database first
obtains the rowid of one of the selected rows by scanning the cluster index. Oracle
Database then locates the rows based on this rowid.
column group
A set of columns that is treated as a unit.
column statistics
Statistics about columns that the optimizer uses to determine optimal execution plans.
Column statistics include the number of distinct column values, low value, high value,
and number of nulls.
concurrency
Simultaneous access of the same data by many users. A multiuser database
management system must provide adequate concurrency controls so that data cannot
be updated or changed improperly, compromising data integrity.
condition
A combination of one or more expressions and logical operators that returns a value of
TRUE, FALSE, or UNKNOWN.
cost
A numeric internal measure that represents the estimated resource usage for an
execution plan. The lower the cost, the more efficient the plan.
Glossary-5
Glossary
cost model
The internal optimizer model that accounts for the cost of the I/O, CPU, and network
resources that a query is predicted to use.
cumulative statistics
A count such as the number of block reads. Oracle Database generates many types of
cumulative statistics for the system, sessions, and individual SQL statements.
cursor
A handle or name for a private SQL area in the PGA. Because cursors are closely
associated with private SQL areas, the terms are sometimes used interchangeably.
cursor cache
See shared SQL area.
cursor merging
Combining cursors to save space in the shared SQL area. If the optimizer creates a
plan for a bind-aware cursor, and if this plan is the same as an existing cursor, then
the optimizer can merge the cursors.
data skew
Large variations in the number of duplicate values in a column.
database operation
A set of database tasks defined by end users or application code, for example, a batch
job or ETL processing.
default plan
For an adaptive plan, the execution plan initially chosen by the optimizer using the
statistics from the data dictionary. The default plan can differ from the final plan.
Glossary-6
Glossary
disabled plan
A plan that a database administrator has manually marked as ineligible for use by the
optimizer.
dense key
A numeric key that is stored as a native integer and has a range of values.
density
A decimal number between 0 and 1 that measures the selectivity of a column. Values
close to 1 indicate that the column is unselective, whereas values close to 0 indicate
that this column is more selective.
driving table
The table to which other tables are joined. An analogy from programming is a for loop
that contains another for loop. The outer for loop is the analog of a driving table, which
is also called an outer table.
dynamic plan
A set of subplan groups. A subplan group is set of subplans. In an adaptive query
plan, the optimizer chooses a subplan at run time depending on the statistics obtained
by the statistics collector.
Glossary-7
Glossary
dynamic statistics
An optimization technique in which the database executes a recursive SQL statement
to scan a small random sample of a table's blocks to estimate predicate selectivities.
enabled plan
In SQL plan management, a plan that is eligible for use by the optimizer.
endpoint number
A number that uniquely identifies a bucket in a histogram. In frequency and hybrid
histograms, the endpoint number is the cumulative frequency of endpoints. In height-
balanced histograms, the endpoint number is the bucket number.
endpoint value
An endpoint value is the highest value in the range of values in a histogram bucket.
equijoin
A join whose join condition contains an equality operator.
estimator
The component of the optimizer that determines the overall cost of a given execution
plan.
execution plan
The combination of steps used by the database to execute a SQL statement. Each
step either retrieves rows of data physically from the database or prepares them for
the session issuing the statement. You can override execution plans by using a hint.
execution tree
A tree diagram that shows the flow of row sources from one step to another in an
execution plan.
expected cardinality
For a table, the estimated number of rows the table has after all filter predicates have
been applied to the table.
Glossary-8
Glossary
expression
A combination of one or more values, operators, and SQL functions that evaluates to a
value. For example, the expression 2*2 evaluates to 4. In general, expressions
assume the data type of their components.
expression statistics
A type of extended statistics that improves optimizer estimates when a WHERE clause
has predicates that use expressions.
extended statistics
A type of optimizer statistics that improves estimates for cardinality when multiple
predicates exist or when predicates contain an expression.
extensible optimizer
An optimizer capability that enables authors of user-defined functions and indexes to
create statistics collection, selectivity, and cost functions that the optimizer uses when
choosing an execution plan. The optimizer cost model is extended to integrate
information supplied by the user to assess CPU and I/O cost.
extension
A column group or an expression. The statistics collected for column groups and
expressions are called extended statistics.
external table
A read-only table whose metadata is stored in the database but whose data in stored
in files outside the database. The database uses the metadata describing external
tables to expose their data as if they were relational tables.
filter condition
A WHERE clause component that eliminates rows from a single object referenced in a
SQL statement.
final plan
In an adaptive plan, the plan that executes to completion. The default plan can differ
from the final plan.
fixed object
A dynamic performance table or its index. The fixed objects are owned by SYS. Fixed
object tables have names beginning with X$ and are the base tables for the V$ views.
fixed plan
An accepted plan that is marked as preferred, so that the optimizer considers only the
fixed plans in the SQL plan baseline. You can use fixed plans to influence the plan
selection process of the optimizer.
Glossary-9
Glossary
frequency histogram
A type of histogram in which each distinct column value corresponds to a single
bucket. An analogy is sorting coins: all pennies go in bucket 1, all nickels go in bucket
2, and so on.
hard parse
The steps performed by the database to build a new executable version of application
code. The database must perform a hard parse instead of a soft parse if the parsed
representation of a submitted statement does not exist in the shared SQL area.
hash cluster
A type of table cluster that is similar to an indexed cluster, except the index key is
replaced with a hash function. No separate cluster index exists. In a hash cluster, the
data is the index.
hash collision
Hashing multiple input values to the same output value.
hash function
A function that operates on an arbitrary-length input value and returns a fixed-length
hash value.
hash join
A method for joining large data sets. The database uses the smaller of two data sets to
build a hash table on the join key in memory. It then scans the larger data set, probing
the hash table to find the joined rows.
hash scan
An access path for a table cluster. The database uses a hash scan to locate rows in a
hash cluster based on a hash value. In a hash cluster, all rows with the same hash
value are stored in the same data block. To perform a hash scan, Oracle Database
Glossary-10
Glossary
first obtains the hash value by applying a hash function to a cluster key value specified
by the statement, and then scans the data blocks containing rows with that hash value.
hash table
An in-memory data structure that associates join keys with rows in a hash join. For
example, in a join of the employees and departments tables, the join key might be the
department ID. A hash function uses the join key to generate a hash value. This hash
value is an index in an array, which is the hash table.
hash value
In a hash cluster, a unique numeric ID that identifies a bucket. Oracle Database uses a
hash function that accepts an infinite number of hash key values as input and sorts
them into a finite number of buckets. Each hash value maps to the database block
address for the block that stores the rows corresponding to the hash key value
(department 10, 20, 30, and so on).
hashing
A mathematical technique in which an infinite set of input values is mapped to a finite
set of output values, called hash values. Hashing is useful for rapid lookups of data in
a hash table.
heap-organized table
A table in which the data rows are stored in no particular order on disk. By default,
CREATE TABLE creates a heap-organized table.
height-balanced histogram
A histogram in which column values are divided into buckets so that each bucket
contains approximately the same number of rows.
hint
An instruction passed to the optimizer through comments in a SQL statement. The
optimizer uses hints to choose an execution plan for the statement.
histogram
A special type of column statistic that provides more detailed information about the
data distribution in a table column.
hybrid histogram
An enhanced height-based histogram that stores the exact frequency of each endpoint
in the sample, and ensures that a value is never stored in multiple buckets.
Glossary-11
Glossary
implicit query
A component of a DML statement that retrieves data without a subquery. An UPDATE,
DELETE, or MERGE statement that does not explicitly include a SELECT statement uses an
implicit query to retrieve the rows to be modified.
In-Memory scan
A table scan that retrieves rows from the In-Memory Column Store (IM column store).
index
Optional schema object associated with a nonclustered table, table partition, or table
cluster. In some cases indexes speed data access.
index cluster
An table cluster that uses an index to locate data. The cluster index is a B-tree index
on the cluster key.
index-organized table
A table whose storage organization is a variant of a primary B-tree index. Unlike a
heap-organized table, data is stored in primary key order.
Glossary-12
Glossary
index statistics
Statistics about indexes that the optimizer uses to determine whether to perform a full
table scan or an index scan. Index statistics include B-tree levels, leaf block counts,
the index clustering factor, distinct keys, and number of rows in the index.
indextype
An object that specifies the routines that manage a domain (application-specific) index.
inner join
A join of two or more tables that returns only those rows that satisfy the join condition.
inner table
In a nested loops join, the table that is not the outer table (driving table). For every
row in the outer table, the database accesses all rows in the inner table. The outer
loop is for every row in the outer table and the inner loop is for every row in the inner
table.
join
A statement that retrieves data from multiple tables specified in the FROM clause of a
SQL statement. Join types include inner joins, outer joins, and Cartesian joins.
join condition
A condition that compares two row sources using an expression. The database
combines pairs of rows, each containing one row from each row source, for which the
join condition evaluates to true.
join elimination
The removal of redundant tables from a query. A table is redundant when its columns
are only referenced in join predicates, and those joins are guaranteed to neither filter
nor expand the resulting rows.
join factorization
A cost-based transformation that can factorize common computations from branches
of a UNION ALL query. Without join factorization, the optimizer evaluates each branch
of a UNION ALL query independently, which leads to repetitive processing, including
data access and joins. Avoiding an extra scan of a large base table can lead to a huge
performance improvement.
Glossary-13
Glossary
join group
A user-created database object that specifies a group of columns that participate in an
join. Join groups are only supported in the In-Memory column store.
join method
A method of joining a pair of row sources. The possible join methods are nested loop,
sort merge, and hash joins. A Cartesian join requires one of the preceding join
methods
join order
The order in which multiple tables are joined together. For example, for each row in
the employees table, the database can read each row in the departments table. In an
alternative join order, for each row in the departments table, the database reads each
row in the employees table.
To execute a statement that joins more than two tables, Oracle Database joins two of
the tables and then joins the resulting row source to the next table. This process
continues until all tables are joined into the result.
join predicate
A predicate in a WHERE or JOIN clause that combines the columns of two tables in a
join.
key vector
A data structure that maps between dense join keys and dense grouping keys.
latch
A low-level serialization control mechanism used to protect shared data structures in
the SGA from simultaneous access.
left table
In an outer join, the table specified on the left side of the OUTER JOIN keywords (in
ANSI SQL syntax).
library cache
An area of memory in the shared pool. This cache includes the shared SQL areas,
private SQL areas (in a shared server configuration), PL/SQL procedures and
packages, and control structures such as locks and library cache handles.
Glossary-14
Glossary
maintenance window
A contiguous time interval during which automated maintenance tasks run. The
maintenance windows are Oracle Scheduler windows that belong to the window group
named MAINTENANCE_WINDOW_GROUP.
materialized view
A schema object that stores a query result. All materialized views are either read-only
or updatable.
multiblock read
An I/O call that reads multiple database blocks. Multiblock reads can significantly
speed up full table scans. For example, a data block might be 8 KB, but the operating
system can read 1024 KB in a single I/O. For some queries, the optimizer may decide
that it is more cost-efficient to read 128 data blocks in one I/O than in 128 sequential
I/Os.
NDV
Number of distinct values. The NDV is important in generating cardinality estimates.
nonequijoin
A join whose join condition does not contain an equality operator.
nonjoin column
A predicate in a WHERE clause that references only one table.
nonpopular value
In a histogram, any value that does not span two or more endpoints. Any value that is
not nonpopular is a popular value.
noworkload statistics
Optimizer system statistics gathered when the database simulates a workload.
Glossary-15
Glossary
optimization
The overall process of choosing the most efficient means of executing a SQL
statement.
optimizer
Built-in database software that determines the most efficient way to execute a SQL
statement by considering factors related to the objects referenced and the conditions
specified in the statement.
optimizer environment
The totality of session settings that can affect execution plan generation, such as the
work area size or optimizer settings (for example, the optimizer mode).
optimizer goal
The prioritization of resource usage by the optimizer. Using the OPTIMIZER_MODE
initialization parameter, you can set the optimizer goal best throughput or best
response time.
optimizer statistics
Details about the database its object used by the optimizer to select the best
execution plan for each SQL statement. Categories include table statistics such as
numbers of rows, index statistics such as B-tree levels, system statistics such as
CPU and I/O performance, and column statistics such as number of nulls.
Glossary-16
Glossary
outer join
A join condition using the outer join operator (+) with one or more columns of one of
the tables. The database returns all rows that meet the join condition. The database
also returns all rows from the table without the outer join operator for which there are
no matching rows in the table with the outer join operator.
outer table
See driving table
parallel execution
The application of multiple CPU and I/O resources to the execution of a single
database operation.
parallel query
A query in which multiple processes work together simultaneously to run a single SQL
query. By dividing the work among multiple processes, Oracle Database can run the
statement more quickly. For example, four processes retrieve rows for four different
quarters in a year instead of one process handling all four quarters by itself.
parent cursor
The cursor that stores the SQL text and other minimal information for a SQL
statement. The child cursor contains the plan, compilation environment, and other
information. When a statement first executes, the database creates both a parent and
child cursor in the shared pool.
parse call
A call to Oracle to prepare a SQL statement for execution. The call includes
syntactically checking the SQL statement, optimizing it, and then building or locating
an executable form of that statement.
parsing
The stage of SQL processing that involves separating the pieces of a SQL statement
into a data structure that can be processed by other routines.
A hard parse occurs when the SQL statement to be executed is either not in the
shared pool, or it is in the shared pool but it cannot be shared. A soft parse occurs
when a session attempts to execute a SQL statement, and the statement is already in
the shared pool, and it can be used.
Glossary-17
Glossary
partition-wise join
A join optimization that divides a large join of two tables, one of which must be
partitioned on the join key, into several smaller joins.
pending statistics
Unpublished optimizer statistics. By default, the optimizer uses published statistics but
does not use pending statistics.
performance feedback
This form of automatic reoptimization helps improve the degree of parallelism
automatically chosen for repeated SQL statements when PARALLEL_DEGREE_POLICY is
set to ADAPTIVE.
plan evolution
The manual change of an unaccepted plan in the SQL plan history into an
accepted plan in the SQL plan baseline.
plan generator
The part of the optimizer that tries different access paths, join methods, and join orders
for a given query block to find the plan with the lowest cost.
plan selection
The attempt to find a matching plan in the SQL plan baseline for a statement after
performing a hard parse.
plan verification
Comparing the performance of an unaccepted plan to a plan in a SQL plan baseline
and ensuring that it performs better.
popular value
In a histogram, any value that spans two or more endpoints. Any value that is not
popular is an nonpopular value.
predicate pushing
A transformation technique in which the optimizer "pushes" the relevant predicates
from the containing query block into the view query block. For views that are not
merged, this technique improves the subplan of the unmerged view because the
database can use the pushed-in predicates to access indexes or to use as filters.
Glossary-18
Glossary
projection view
An optimizer-generated view that appear in queries in which a DISTINCT view has
been merged, or a GROUP BY view is merged into an outer query block that also
contains GROUP BY, HAVING, or aggregates.
query
An operation that retrieves data from tables or views. For example, SELECT * FROM
employees is a query.
query block
A top-level SELECT statement, subquery, or unmerged view
query optimizer
See optimizer.
recursive SQL
Additional SQL statements that the database must issue to execute a SQL statement
issued by a user. The generation of recursive SQL is known as a recursive call. For
example, the database generates recursive calls when data dictionary information is
not available in memory and so must be retrieved from disk.
reoptimization
See automatic reoptimization.
response time
The time required to complete a unit of work.
Glossary-19
Glossary
See throughput.
result set
In a query, the set of rows generated by the execution of a cursor.
right table
In an outer join, the table specified on the right side of the OUTER JOIN keywords (in
ANSI SQL syntax).
rowid
A globally unique address for a row in a table.
row set
A set of rows returned by a step in an execution plan.
row source
An iterative control structure that processes a set of rows in an iterated manner and
produces a row set.
rule filter
The use of DBMS_STATS.CONFIGURE_ADVISOR_RULE_FILTER to restrict an Optimizer
Statistics Advisor task to a user-specified set of rules. For example, you might exclude
the rule that checks for stale statistics.
sampling
Gathering statistics from a random subset of rows in a table.
selectivity
A value indicating the proportion of a row set retrieved by a predicate or combination
of predicates, for example, WHERE last_name = 'Smith'. A selectivity of 0 means that
Glossary-20
Glossary
no rows pass the predicate test, whereas a value of 1 means that all rows pass the
test.
The adjective selective means roughly "choosy." Thus, a highly selective query returns
a low proportion of rows (selectivity close to 0), whereas an unselective query returns
a high proportion of rows (selectivity close to 1).
semijoin
A join that returns rows from the first table when at least one match exists in the
second table. For example, you list departments with at least one employee. The
difference between a semijoin and a conventional join is that rows in the first table are
returned at most once. Semijoins use the EXISTS or IN constructs.
shared cursor
A shared SQL area that is used by multiple SQL statements.
shared pool
Portion of the SGA that contains shared memory constructs such as shared SQL
areas.
SMB
See SQL management base (SMB).
snowflake schema
A star schema in which dimension tables reference other tables.
snowstorm schema
A combination of multiple snowflake schemas.
soft parse
Any parse that is not a hard parse. If a submitted SQL statement is the same as a
reusable SQL statement in the shared pool, then Oracle Database reuses the existing
code. This reuse of code is also called a library cache hit.
Glossary-21
Glossary
SQL compilation
In the context of Oracle SQL processing, this term refers collectively to the phases of
parsing, optimization, and plan generation.
SQL handle
A string value derived from the numeric SQL signature. Like the signature, the handle
uniquely identifies a SQL statement. It serves as a SQL search key in user APIs.
SQL ID
For a specific SQL statement, the unique identifier of the parent cursor in the library
cache. A hash function applied to the text of the SQL statement generates the SQL ID.
The V$SQL.SQL_ID column displays the SQL ID.
SQL incident
In the fault diagnosability infrastructure of Oracle Database, a single occurrence of a
SQL-related problem. When a problem (critical error) occurs multiple times, the
database creates an incident for each occurrence. Incidents are timestamped and
tracked in the Automatic Diagnostic Repository (ADR).
Glossary-22
Glossary
SQL processing
The stages of parsing, optimization, row source generation, and execution of a SQL
statement.
SQL profile
A set of auxiliary information built during automatic tuning of a SQL statement. A SQL
profile is to a SQL statement what statistics are to a table. The optimizer can use SQL
profiles to improve cardinality and selectivity estimates, which in turn leads the
optimizer to select better plans.
SQL profiling
The verification and validation by the Automatic Tuning Advisor of its own estimates.
SQL signature
A numeric hash value computed using a SQL statement text that has been normalized
for case insensitivity and white space. It uniquely identifies a SQL statement. The
database uses this signature as a key to maintain SQL management objects such as
SQL profiles, SQL plan baselines, and SQL patches.
Glossary-23
Glossary
CPU and elapsed times, physical reads and logical reads, and misses in the library
cache.
SQL tuning
The process of improving SQL statement efficiency to meet measurable goals.
star schema
A relational schema whose design represents a dimensional data model. The star
schema consists of one or more fact tables and one or more dimension tables that are
related through foreign keys.
statistics feedback
A form of automatic reoptimization that automatically improves plans for repeated
queries that have cardinality misestimates. The optimizer may estimate cardinalities
incorrectly for many reasons, such as missing statistics, inaccurate statistics, or
complex predicates.
stored outline
A set of hints for a SQL statement. The hints in stored outlines direct the optimizer to
choose a specific plan for the statement.
subplan
A portion of an adaptive plan that the optimizer can switch to as an alternative at run
time. A subplan can consist of multiple operations in the plan. For example, the
optimizer can treat a join method and the corresponding access path as one unit when
determining whether to change the plan at run time.
subplan group
A set of subplans in an adaptive query plan.
subquery
A query nested within another SQL statement. Unlike implicit queries, subqueries use
a SELECT statement to retrieve data.
subquery unnesting
A transformation technique in which the optimizer transforms a nested query into an
equivalent join statement, and then optimizes the join.
Glossary-24
Glossary
synopsis
A set of auxiliary statistics gathered on a partitioned table when the INCREMENTAL value
is set to true.
system statistics
Statistics that enable the optimizer to use CPU and I/O characteristics. Index statistics
include B-tree levels, leaf block counts, clustering factor, distinct keys, and number of
rows in the index.
table cluster
A schema object that contains data from one or more tables, all of which have one or
more columns in common. In table clusters, the database stores together all the rows
from all tables that share the same cluster key.
table expansion
A transformation technique that enables the optimizer to generate a plan that uses
indexes on the read-mostly portion of a partitioned table, but not on the active portion
of the table.
table statistics
Statistics about tables that the optimizer uses to determine table access cost, join
cardinality, join order, and so on. Table statistics include row counts, block counts,
empty blocks, average free space per block, number of chained rows, average row
length, and staleness of the statistics on the table.
throughput
The amount of work completed in a unit of time.
tuning mode
One of the two optimizer modes. When running in tuning mode, the optimizer is known
as the Automatic Tuning Optimizer. In tuning mode, the optimizer determines
whether it can further improve the plan produced in normal mode. The optimizer output
is not an execution plan, but a series of actions, along with their rationale and
expected benefit for producing a significantly better plan.
unaccepted plan
A plan for a statement that is in the SQL plan history but has not been added to the
SQL plan management.
unselective
A relatively large fraction of rows from a row set. A query becomes more unselective
as the selectivity approaches 1. For example, a query that returns 999,999 rows from
Glossary-25
Glossary
a table with one million rows is unselective. A query of the same table that returns one
row is selective.
See throughput.
V$ view
See dynamic performance view.
vector I/O
A type of I/O in which the database obtains a set of rowids, sends them batched in an
array to the operating system, which performs the read.
view merging
The merging of a query block representing a view into the query block that contains it.
View merging can improve plans by enabling the optimizer to consider additional join
orders, access methods, and other transformations.
workload statistics
Optimizer statistics for system activity in a specified time period.
Glossary-26
Index
A advisors
Optimizer Statistics Advisor, 18-1–18-3
accepted plans, 4-31 SPM Evolve Advisor, 29-5, 29-10
access paths, 3-8, 8-1 SQL Access Advisor, 26-1, 26-8
B-tree indexes, 8-15 SQL Tuning Advisor, 25-1
bitmap index, 8-37 ALTER SESSION statement, 23-14
bitmap index range scans, 8-45 ANALYZE command, 16-2
bitmap indexes, 8-37, 8-38, 8-47 antijoins, 9-4, 9-42
execution plans, 6-1 handling nulls, 9-45
full table scans, 8-6, 10-6, 11-2 how they work, 9-44
heap-organized tables, 8-3 when the optimizer considers, 9-43
In-Memory table scans, 8-12, 8-14 APPEND hint, 10-16
index fast full, 8-30 applications
index full scans, 8-28 implementing, 2-3
index join scans, 8-34 approximate query processing, 4-29
index range scans, 8-24 automatic plan capture
index skip scans, 8-32 creating automatic capture filters, 29-8
index unique scans, 8-21 enabling, 29-6
optimizing with SQL Access Advisor, 26-10 automatic reoptimization, 4-14, 4-24, 7-3, 10-24
sample table scans, 8-11, 8-12 cardinality misestimates, 4-25
table, 8-2 performance feedback, 4-27
table access by rowid, 8-9 statistics feedback, 4-25
table cluster, 8-49 automatic statistics collection, 13-2
adaptive query optimization, 4-14 Automatic Tuning Optimizer, 1-6, 4-13
adaptive plans, 22-2 SQL profiles, 25-7
adaptive query plans, 4-15, 7-3, 7-32 Automatic Workload Repository (AWR), 1-6
controlling, 19-10
dynamic statistics, 10-15
adaptive query plans, 4-15, 7-3, 7-32, 22-2
B
about, 4-15 B-tree indexes, 8-15
bitmap index pruning, 4-22 band joins, 9-30
cardinality misestimates, 4-16 big bang rollout strategy, 2-4
enabling, 4-23 bind variables, 20-1, 20-2
join methods, 4-17 bind peeking, 20-15
optimizer statistics collector, 4-16 cursors, 20-13
parallel distribution methods, 4-20 substitution, 20-22
subplans, 4-16 bind-aware cursors, 20-32
adaptive sampling, 13-26 bind-sensitive cursors, 20-28
adaptive statistics, 4-24 bitmap indexes
automatic reoptimization, 4-24 access paths, 8-37
dynamic statistics, 4-24 bitmap merge, 8-47
SQL plan directives, 4-28, 14-1 conversion to rowid, 8-42
when enabled, 4-29 inlist iterator, 7-20
ADDM, 1-6 on joins, A-12
pruning, 4-22
Index-1
Index
Index-2
Index
3
Index
Index-4
Index
5
Index
Index-6
Index
7
Index
Index-8
Index
9
Index
Index-10
Index
11
Index
Index-12